Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity
Identifieur interne : 000648 ( Main/Corpus ); précédent : 000647; suivant : 000649Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity
Auteurs : Hayriye Cagnan ; Kevin Dolan ; Xuan He ; Maria Fiorella Contarino ; Richard Schuurman ; Pepijn Van Den Munckhof ; Wytse J. Wadman ; Lo Bour ; Hubert C F. MartensSource :
- Journal of Neural Engineering [ 1741-2560 ] ; 2011.
Abstract
Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.
Url:
Links to Exploration step
ISTEX:AE48434E0788CA8C26BA50FD02CDF77DA015C6C1Le document en format XML
<record><TEI wicri:istexFullTextTei="biblStruct"><teiHeader><fileDesc><titleStmt><title>Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title><author><name sortKey="Cagnan, Hayriye" sort="Cagnan, Hayriye" uniqKey="Cagnan H" first="Hayriye" last="Cagnan">Hayriye Cagnan</name><affiliation><mods:affiliation>Department of Clinical Neurology, University of Oxford, UK</mods:affiliation></affiliation><affiliation><mods:affiliation>Author to whom any correspondence should be addressed</mods:affiliation></affiliation><affiliation><mods:affiliation>E-mail:hayriye.cagnan@clneuro.ox.ac.uk</mods:affiliation></affiliation></author><author><name sortKey="Dolan, Kevin" sort="Dolan, Kevin" uniqKey="Dolan K" first="Kevin" last="Dolan">Kevin Dolan</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="He, Xuan" sort="He, Xuan" uniqKey="He X" first="Xuan" last="He">Xuan He</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Contarino, Maria Fiorella" sort="Contarino, Maria Fiorella" uniqKey="Contarino M" first="Maria Fiorella" last="Contarino">Maria Fiorella Contarino</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Schuurman, Richard" sort="Schuurman, Richard" uniqKey="Schuurman R" first="Richard" last="Schuurman">Richard Schuurman</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Van Den Munckhof, Pepijn" sort="Van Den Munckhof, Pepijn" uniqKey="Van Den Munckhof P" first="Pepijn" last="Van Den Munckhof">Pepijn Van Den Munckhof</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Wadman, Wytse J" sort="Wadman, Wytse J" uniqKey="Wadman W" first="Wytse J" last="Wadman">Wytse J. Wadman</name><affiliation><mods:affiliation>SILS, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Bour, Lo" sort="Bour, Lo" uniqKey="Bour L" first="Lo" last="Bour">Lo Bour</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Martens, Hubert C F" sort="Martens, Hubert C F" uniqKey="Martens H" first="Hubert C F" last="Martens">Hubert C F. Martens</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author></titleStmt><publicationStmt><idno type="wicri:source">ISTEX</idno><idno type="RBID">ISTEX:AE48434E0788CA8C26BA50FD02CDF77DA015C6C1</idno><date when="2011" year="2011">2011</date><idno type="url">https://api.istex.fr/document/AE48434E0788CA8C26BA50FD02CDF77DA015C6C1/fulltext/pdf</idno><idno type="wicri:Area/Main/Corpus">000648</idno></publicationStmt><sourceDesc><biblStruct><analytic><title level="a">Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title><author><name sortKey="Cagnan, Hayriye" sort="Cagnan, Hayriye" uniqKey="Cagnan H" first="Hayriye" last="Cagnan">Hayriye Cagnan</name><affiliation><mods:affiliation>Department of Clinical Neurology, University of Oxford, UK</mods:affiliation></affiliation><affiliation><mods:affiliation>Author to whom any correspondence should be addressed</mods:affiliation></affiliation><affiliation><mods:affiliation>E-mail:hayriye.cagnan@clneuro.ox.ac.uk</mods:affiliation></affiliation></author><author><name sortKey="Dolan, Kevin" sort="Dolan, Kevin" uniqKey="Dolan K" first="Kevin" last="Dolan">Kevin Dolan</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="He, Xuan" sort="He, Xuan" uniqKey="He X" first="Xuan" last="He">Xuan He</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Contarino, Maria Fiorella" sort="Contarino, Maria Fiorella" uniqKey="Contarino M" first="Maria Fiorella" last="Contarino">Maria Fiorella Contarino</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Schuurman, Richard" sort="Schuurman, Richard" uniqKey="Schuurman R" first="Richard" last="Schuurman">Richard Schuurman</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Van Den Munckhof, Pepijn" sort="Van Den Munckhof, Pepijn" uniqKey="Van Den Munckhof P" first="Pepijn" last="Van Den Munckhof">Pepijn Van Den Munckhof</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Wadman, Wytse J" sort="Wadman, Wytse J" uniqKey="Wadman W" first="Wytse J" last="Wadman">Wytse J. Wadman</name><affiliation><mods:affiliation>SILS, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Bour, Lo" sort="Bour, Lo" uniqKey="Bour L" first="Lo" last="Bour">Lo Bour</name><affiliation><mods:affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</mods:affiliation></affiliation></author><author><name sortKey="Martens, Hubert C F" sort="Martens, Hubert C F" uniqKey="Martens H" first="Hubert C F" last="Martens">Hubert C F. Martens</name><affiliation><mods:affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</mods:affiliation></affiliation></author></analytic><monogr></monogr><series><title level="j">Journal of Neural Engineering</title><title level="j" type="abbrev">J. Neural Eng.</title><idno type="ISSN">1741-2560</idno><imprint><publisher>IOP Publishing</publisher><date type="published" when="2011">2011</date><biblScope unit="volume">8</biblScope><biblScope unit="issue">4</biblScope><biblScope unit="page" from="1">1</biblScope><biblScope unit="page" to="9">9</biblScope><biblScope unit="production">Printed in the UK</biblScope></imprint></series><idno type="istex">AE48434E0788CA8C26BA50FD02CDF77DA015C6C1</idno><idno type="DOI">10.1088/1741-2560/8/4/046006</idno><idno type="PII">S1741-2560(11)85116-5</idno><idno type="articleID">385116</idno><idno type="articleNumber">046006</idno></biblStruct></sourceDesc></fileDesc><profileDesc><textClass></textClass><langUsage><language ident="en">en</language></langUsage></profileDesc></teiHeader><front><div type="abstract">Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.</div></front></TEI><istex><corpusName>iop</corpusName><author><json:item><name>Hayriye Cagnan</name><affiliations><json:string>Department of Clinical Neurology, University of Oxford, UK</json:string><json:string>Author to whom any correspondence should be addressed</json:string><json:string>E-mail:hayriye.cagnan@clneuro.ox.ac.uk</json:string></affiliations></json:item><json:item><name>Kevin Dolan</name><affiliations><json:string>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</json:string></affiliations></json:item><json:item><name>Xuan He</name><affiliations><json:string>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</json:string></affiliations></json:item><json:item><name>Maria Fiorella Contarino</name><affiliations><json:string>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</json:string></affiliations></json:item><json:item><name>Richard Schuurman</name><affiliations><json:string>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</json:string></affiliations></json:item><json:item><name>Pepijn van den Munckhof</name><affiliations><json:string>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</json:string></affiliations></json:item><json:item><name>Wytse J Wadman</name><affiliations><json:string>SILS, Universiteit van Amsterdam, The Netherlands</json:string></affiliations></json:item><json:item><name>Lo Bour</name><affiliations><json:string>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</json:string></affiliations></json:item><json:item><name>Hubert C F Martens</name><affiliations><json:string>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</json:string></affiliations></json:item></author><subject><json:item><lang><json:string>eng</json:string></lang><value>Parkinson's disease</value></json:item><json:item><lang><json:string>eng</json:string></lang><value>deep brain stimulation</value></json:item><json:item><lang><json:string>eng</json:string></lang><value>brain mapping</value></json:item></subject><language><json:string>eng</json:string></language><abstract>Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.</abstract><qualityIndicators><score>8.452</score><pdfVersion>1.4</pdfVersion><pdfPageSize>595 x 842 pts (A4)</pdfPageSize><refBibsNative>true</refBibsNative><keywordCount>3</keywordCount><abstractCharCount>1667</abstractCharCount><pdfWordCount>6005</pdfWordCount><pdfCharCount>36719</pdfCharCount><pdfPageCount>9</pdfPageCount><abstractWordCount>246</abstractWordCount></qualityIndicators><title>Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title><pii><json:string>S1741-2560(11)85116-5</json:string></pii><genre><json:string>article</json:string></genre><host><volume>8</volume><pages><total>9</total><last>9</last><first>1</first></pages><issn><json:string>1741-2560</json:string></issn><issue>4</issue><genre></genre><language><json:string>unknown</json:string></language><title>Journal of Neural Engineering</title></host><publicationDate>2011</publicationDate><copyrightDate>2011</copyrightDate><doi><json:string>10.1088/1741-2560/8/4/046006</json:string></doi><id>AE48434E0788CA8C26BA50FD02CDF77DA015C6C1</id><fulltext><json:item><original>true</original><mimetype>application/pdf</mimetype><extension>pdf</extension><uri>https://api.istex.fr/document/AE48434E0788CA8C26BA50FD02CDF77DA015C6C1/fulltext/pdf</uri></json:item><json:item><original>false</original><mimetype>application/zip</mimetype><extension>zip</extension><uri>https://api.istex.fr/document/AE48434E0788CA8C26BA50FD02CDF77DA015C6C1/fulltext/zip</uri></json:item><istex:fulltextTEI uri="https://api.istex.fr/document/AE48434E0788CA8C26BA50FD02CDF77DA015C6C1/fulltext/tei"><teiHeader><fileDesc><titleStmt><title level="a">Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title></titleStmt><publicationStmt><authority>ISTEX</authority><publisher>IOP Publishing</publisher><availability><p>2011 IOP Publishing Ltd</p></availability><date>2011</date></publicationStmt><sourceDesc><biblStruct type="inbook"><analytic><title level="a">Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title><author><persName><forename type="first">Hayriye</forename><surname>Cagnan</surname></persName><affiliation>Department of Clinical Neurology, University of Oxford, UK</affiliation><affiliation>Author to whom any correspondence should be addressed</affiliation><affiliation>E-mail:hayriye.cagnan@clneuro.ox.ac.uk</affiliation></author><author><persName><forename type="first">Kevin</forename><surname>Dolan</surname></persName><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation></author><author><persName><forename type="first">Xuan</forename><surname>He</surname></persName><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation></author><author><persName><forename type="first">Maria Fiorella</forename><surname>Contarino</surname></persName><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation></author><author><persName><forename type="first">Richard</forename><surname>Schuurman</surname></persName><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation></author><author><persName><forename type="first">Pepijn</forename><surname>van den Munckhof</surname></persName><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation></author><author><persName><forename type="first">Wytse J</forename><surname>Wadman</surname></persName><affiliation>SILS, Universiteit van Amsterdam, The Netherlands</affiliation></author><author><persName><forename type="first">Lo</forename><surname>Bour</surname></persName><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation></author><author><persName><forename type="first">Hubert C F</forename><surname>Martens</surname></persName><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation></author></analytic><monogr><title level="j">Journal of Neural Engineering</title><title level="j" type="abbrev">J. Neural Eng.</title><idno type="pISSN">1741-2560</idno><imprint><publisher>IOP Publishing</publisher><date type="published" when="2011"></date><biblScope unit="volume">8</biblScope><biblScope unit="issue">4</biblScope><biblScope unit="page" from="1">1</biblScope><biblScope unit="page" to="9">9</biblScope><biblScope unit="production">Printed in the UK</biblScope></imprint></monogr><idno type="istex">AE48434E0788CA8C26BA50FD02CDF77DA015C6C1</idno><idno type="DOI">10.1088/1741-2560/8/4/046006</idno><idno type="PII">S1741-2560(11)85116-5</idno><idno type="articleID">385116</idno><idno type="articleNumber">046006</idno></biblStruct></sourceDesc></fileDesc><profileDesc><creation><date>2011</date></creation><langUsage><language ident="en">en</language></langUsage><abstract><p>Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.</p></abstract><textClass><keywords scheme="keyword"><list><head>Keywords</head><item><term>Parkinson's disease</term></item><item><term>deep brain stimulation</term></item><item><term>brain mapping</term></item></list></keywords></textClass></profileDesc><revisionDesc><change when="2011">Published</change></revisionDesc></teiHeader></istex:fulltextTEI><json:item><original>false</original><mimetype>text/plain</mimetype><extension>txt</extension><uri>https://api.istex.fr/document/AE48434E0788CA8C26BA50FD02CDF77DA015C6C1/fulltext/txt</uri></json:item></fulltext><metadata><istex:metadataXml wicri:clean="corpus iop not found" wicri:toSee="no header"><istex:xmlDeclaration>version="1.0" encoding="ISO-8859-1" </istex:xmlDeclaration><istex:docType SYSTEM="http://ej.iop.org/dtd/iopv1_5_2.dtd" name="istex:docType"></istex:docType><istex:document><article artid="jne385116"><article-metadata><jnl-data jnlid="jne"><jnl-fullname>Journal of Neural Engineering</jnl-fullname><jnl-abbreviation>J. Neural Eng.</jnl-abbreviation><jnl-shortname>JNE</jnl-shortname><jnl-issn>1741-2560</jnl-issn><jnl-coden>JNEIEZ</jnl-coden><jnl-imprint>IOP Publishing</jnl-imprint><jnl-web-address>stacks.iop.org/JNE</jnl-web-address></jnl-data><volume-data><year-publication>2011</year-publication><volume-number>8</volume-number></volume-data><issue-data><issue-number>4</issue-number><coverdate>August 2011</coverdate></issue-data><article-data><article-type type="paper" sort="regular"></article-type><type-number numbering="article" type="paper" artnum="046006">046006</type-number><article-number>385116</article-number><first-page>1</first-page><last-page>9</last-page><length>9</length><pii>S1741-2560(11)85116-5</pii><doi>10.1088/1741-2560/8/4/046006</doi><copyright>2011 IOP Publishing Ltd</copyright><ccc>1741-2560/11/046006+09$33.00</ccc><printed>Printed in the UK</printed><features colour="global" mmedia="yes" suppdata="yes"></features></article-data></article-metadata><header><title-group><title>Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title><short-title>Automatic subthalamic nucleus detection</short-title><ej-title>Automatic subthalamic nucleus detection</ej-title></title-group><author-group><author address="jne385116ad1" alt-address="jne385116ad5" email="jne385116ea1"><first-names>Hayriye</first-names><second-name>Cagnan</second-name></author><author address="jne385116ad2"><first-names>Kevin</first-names><second-name>Dolan</second-name></author><author address="jne385116ad2"><first-names>Xuan</first-names><second-name>He</second-name></author><author address="jne385116ad3"><first-names>Maria Fiorella</first-names><second-name>Contarino</second-name></author><author address="jne385116ad3"><first-names>Richard</first-names><second-name>Schuurman</second-name></author><author address="jne385116ad3"><first-names>Pepijn</first-names><second-name>van den Munckhof</second-name></author><author address="jne385116ad4"><first-names>Wytse J</first-names><second-name>Wadman</second-name></author><author address="jne385116ad3"><first-names>Lo</first-names><second-name>Bour</second-name></author><author address="jne385116ad2"><first-names>Hubert C F</first-names><second-name>Martens</second-name></author></author-group><address-group><address id="jne385116ad1"><orgname>Department of Clinical Neurology, University of Oxford</orgname>, <country>UK</country></address><address id="jne385116ad2"><orgname>Philips Research Laboratories</orgname>, Eindhoven, 5656 AE, <country>The Netherlands</country></address><address id="jne385116ad3"><orgname>Academic Medical Center, Universiteit van Amsterdam</orgname>, <country>The Netherlands</country></address><address id="jne385116ad4"><orgname>SILS, Universiteit van Amsterdam</orgname>, <country>The Netherlands</country></address><address id="jne385116ad5" alt="yes">Author to whom any correspondence should be addressed</address><e-address id="jne385116ea1"><email mailto="hayriye.cagnan@clneuro.ox.ac.uk">hayriye.cagnan@clneuro.ox.ac.uk</email></e-address></address-group><history received="18 February 2011" accepted="5 May 2011" online="31 May 2011"></history><abstract-group><abstract><heading>Abstract</heading><p indent="no">Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88%. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.</p></abstract></abstract-group><classifications><keywords><keyword>Parkinson's disease</keyword><keyword>deep brain stimulation</keyword><keyword>brain mapping</keyword></keywords></classifications></header><body numbering="bysection"><sec-level1 id="jne385116s1" label="1"><heading>Introduction</heading><p indent="no">Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a widely used surgical technique in the management of late stage Parkinson's disease (PD) motor symptoms. Amongst others, the efficacy of DBS is dependent on accurate localization of the target nucleus [<cite linkend="jne385116bib01" range="jne385116bib01,jne385116bib02,jne385116bib03,jne385116bib04">1–4</cite>]. Therefore, during DBS surgery, microelectrode recording (MER) of neuronal activity along the surgical trajectory is often performed to refine target location [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib03" range="jne385116bib03,jne385116bib04,jne385116bib05,jne385116bib06,jne385116bib07,jne385116bib08,jne385116bib09,jne385116bib10">3–10</cite>]. The main aim of MER mapping is accurate delineation of the functional boundaries of the STN and its surrounding structures [<cite linkend="jne385116bib01" range="jne385116bib01,jne385116bib02,jne385116bib03,jne385116bib04,jne385116bib05">1–5</cite>, <cite linkend="jne385116bib07">7</cite>, <cite linkend="jne385116bib11">11</cite>]. Typically, audio and visual conversions of the MERs are monitored and assessed by experts during surgery [<cite linkend="jne385116bib05">5</cite>]. Based on characteristic MER patterns, functional targets are identified. Dorsal to the STN, the thalamus is passed showing relatively slowly firing and bursting activity patterns [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib07">7</cite>, <cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15">12–15</cite>]. In PD, the hyper-activity of the STN neurons is reflected as increased background noise level, high firing rate and irregular bursting activity patterns [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib06">6</cite>, <cite linkend="jne385116bib07">7</cite>, <cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15">12–15</cite>]. Ventral to the STN, the <italic>substantia nigra pars reticulata</italic> (SNr) may be encountered, which is characterized by more regularly firing units [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib07">7</cite>]. Correct interpretation of the MERs is important for targeting but requires time and expertise and can be challenging especially under surgical circumstances. As the number of centers performing DBS surgeries continues to grow, there is a need for a system that can support surgical teams in real time with reliable and objective identification of the target nucleus from the MERs.</p><p>Several studies have addressed MER-based automatic localization and visualization of the STN [<cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15,jne385116bib16">12–16</cite>]. Commonly utilized signal features are the background noise level and spike count [<cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15,jne385116bib16">12–16</cite>]. Since synchronized oscillatory activity patterns in the theta and beta frequency bands seem to be characteristic in PD patients, power spectral density (PSD)-based measures also have been used for STN detection and even for discrimination of the sub-territories of the STN [<cite linkend="jne385116bib03">3</cite>, <cite linkend="jne385116bib13">13</cite>, <cite linkend="jne385116bib15">15</cite>, <cite linkend="jne385116bib17" range="jne385116bib17,jne385116bib18,jne385116bib19,jne385116bib20,jne385116bib21,jne385116bib22">17–22</cite>]. Studies addressing MER-based automatic localization and visualization of the STN have generally relied on a single feature in order to distinguish the STN from the neighboring structures [<cite linkend="jne385116bib14">14</cite>, <cite linkend="jne385116bib16">16</cite>]. Zaidel <italic>et al</italic> made use of changes in the background noise level in order to detect the STN and combined the background noise level with PSD-based measures in order to detect oscillatory sub-territories within the STN [<cite linkend="jne385116bib15">15</cite>]. In order to account for inter-subject variability, Wong <italic>et al</italic> have employed 13 measures, characterizing spiking activity and noise, in an algorithm used for visualization of the different regions encountered along the surgical trajectory [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib12">12</cite>]. Measures used for visualizing and detecting the STN are based on years of clinical experience on MERs [<cite linkend="jne385116bib01">1</cite>]. While commonly used in various studies, to our knowledge, the predictive value and selectivity of these measures for delineation of the STN borders have not been investigated. Due to inter-subject variability as a result of different disease states or prior medical history of the patients, changes in the MER features specific to the STN can show variation. Moreover, the presence of artifacts can modify the reliability of the MER features. One potential approach, which is highlighted in this study, is to use multiple features in the detection of the STN borders and to indicate the reliability of the annotations made automatically based on the predictive value and selectivity of the features used.</p><p>The main objective of this study is the development and validation of an unassisted algorithm that identifies the STN and the SNr from surgical MERs. To this end, we firstly assess the predictive value of the background noise level, firing rate and PSD for delineation of the STN borders. We then construct an algorithm to automatically identify the STN and the SNr using these measures. The algorithm detects the dorsal and ventral borders of the STN using different combinations of the measures and allocates qualitative confidence levels (i.e. high, medium, low) based on combinations of measures used. To test the algorithm and assess its reliability, we have applied it to a large number of consecutive recordings without any pre-selection of the data and compared its results to the surgical annotations and to independently made offline annotations by two MER experts.</p></sec-level1><sec-level1 id="jne385116s2" label="2"><heading>Patients and methods</heading><sec-level2 id="jne385116s2-1" label="2.1"><heading>MERs, surgical and expert annotations</heading><p indent="no">Over a period of 4 years, 48 PD patients received DBS in the STN (84 electrode implantations) at the Academic Medical Center of Amsterdam. For each hemisphere, one to five MER trajectories were performed (figure <figref linkend="jne385116fig01">1</figref>). MERs were collected by the Department of Neurology/Clinical Neurophysiology of the Academic Medical Center of Amsterdam and were acquired using the Leadpoint™ system (Medtronic Inc.) [<cite linkend="jne385116bib05">5</cite>]. Details about the surgical procedure and MER acquisition are provided as supplementary material (SM1) available at <stacks mmedia="yes"></stacks>. Anonymized MERs (258 MER trajectories, 6064 recording sites) were retrospectively analyzed for this study.
<figure id="jne385116fig01" width="page"><graphic><graphic-file version="print" format="EPS" filename="images/jne385116fig01.eps" width="36pc"></graphic-file><graphic-file version="ej" format="JPEG" filename="images/jne385116fig01.jpg"></graphic-file></graphic><caption id="jne385116fc01" label="Figure 1"><p indent="no">10 s MER epochs obtained at different depths. Depth values are noted with respect to the pre-operative MRI-based target and 0 mm indicates the center of the STN. Recordings were made at every 0.5 mm. Recordings indicated in blue indicate regions outside of the STN; in red, recordings made from inside the STN are shown; and in green, recordings made inside the SNr are indicated, according to the surgical annotations made.</p></caption></figure></p><p>For the 84 STN implantations, annotations made during the DBS surgery were retrieved. These surgical annotations indicated the dorsal and ventral boundaries of the STN as well as the entry to the SNr. In addition, for a set of 42 randomly chosen trajectories expert annotations were independently created off-line by two MER experts (LB and MFC), blinded to the surgical annotations. These 42 trajectories have been used as training data sets to optimize the algorithm.</p></sec-level2><sec-level2 id="jne385116s2-2" label="2.2"><heading>Automatic STN detection algorithm</heading><sec-level3 id="jne385116s2-2-1" label="2.2.1"><heading>Automatic MER signal processing</heading><p indent="no">Prior to automatic localization of the STN and the SNr, signal processing is used for delineation of noise, artifacts and spikes in the MER. First, the noise level is estimated using the envelope of the MER (figures <figref linkend="jne385116fig02" override="yes">2(A)</figref> and (B)) [<cite linkend="jne385116bib23">23</cite>]. The estimated noise level is used to detect high amplitude artifacts and frequency spectrum of the MER is used to detect low amplitude artifacts (figure <figref linkend="jne385116fig02" override="yes">2(A)</figref>). Automatically detected artifacts are removed from the MER and excluded from further analysis. Sections of the MER exceeding an amplitude threshold, which is based on the estimated noise level, are compared to a spike template and are marked either as spikes or as low amplitude artifacts (figure <figref linkend="jne385116fig02" override="yes">2(B)</figref>). The spike template used in this study is a general template and can be used for spike detection along the entire surgical trajectory. Sections of the MER, not meeting the spike criteria and subsequently marked as low amplitude artifacts, are also excluded from further analysis. Further particulars on noise level estimation, artifact removal and spike detection can be found in the following sections.
<figure id="jne385116fig02"><graphic><graphic-file version="print" format="EPS" filename="images/jne385116fig02.eps" width="18pc"></graphic-file><graphic-file version="ej" format="JPEG" filename="images/jne385116fig02.jpg"></graphic-file></graphic><caption id="jne385116fc02" label="Figure 2"><p indent="no">(A) Example of 5 s MER epoch containing mechanical artifacts automatically detected by the software (in magenta). Background signal noise is given in blue. (B) Example of 50 ms MER epoch containing automatically detected spiking activity (in cyan).</p></caption></figure></p><p indent="no"><italic>Noise level estimation</italic>. Noise level is estimated using the envelope of the MER signal [<cite linkend="jne385116bib23">23</cite>]. The envelope of the signal is estimated as follows. First, the Hilbert transform of the original signal <italic>x</italic>(<italic>t</italic>) is taken:
<display-eqn id="jne385116ueq01" number="no"></display-eqn>Then, an analytical signal is constructed whose real part is <italic>x</italic>(<italic>t</italic>) and imaginary part is <italic>H</italic>(<italic>x</italic>(<italic>t</italic>)). The instantaneous amplitude of this analytical signal is an estimate of the signal envelope [<cite linkend="jne385116bib23">23</cite>]. For Gaussian distributed noise, the distribution of the instantaneous amplitudes of the signal is equivalent to the Rayleigh distribution, where the distribution's mode corresponds to the standard deviation of the background noise. For realistic signals encountered during DBS surgery (i.e. containing high frequency firing and artifacts) the mode proves a robust statistic and is used as our noise level estimate [<cite linkend="jne385116bib23">23</cite>].</p><p indent="no"><italic>Artifact detection</italic>. Mechanical artifacts are common in MER and may lead to misinterpretation of the data. Therefore, prior to spike detection, artifacts are removed automatically from the signal in order to prevent spike-dependent statistics (i.e. compound firing rate) and spectral measures from being biased. Artifacts are detected automatically using a combined amplitude and frequency criterion relative to the estimated noise level; any artifact contaminated data segment is suppressed from further analysis (figure <figref linkend="jne385116fig02" override="yes">2(A)</figref>). First, high amplitude artifacts are detected by marking any data segment of amplitude 7 times the estimated noise level as containing artifacts. For the frequency analysis, the MER is divided into 50 ms windows and Fourier analysis is applied to each window. If the maximum amplitude of the Fourier transform of the 50 ms window is 2.5 times higher than the median maximum amplitude of the Fourier transform of the preceding windows, then the window is marked as containing artifacts. Artifacts that may remain in the data are subsequently rejected during spike detection. The maximum amplitude of the Fourier transform of a 50 ms window varies depending on the content of the 50 ms window: noise, noise and sparse spiking, noise and high frequency spiking, noise and artifact (in ascending order). Specifically, the differentiation between high frequency spiking and artifact relies on the existence of noise in between two spikes due to the refractory period of the neuron versus absence of noise embedded within a mechanical artifact. Amplitude criteria (i.e. 7 times the estimated noise level) and frequency criteria (i.e. 2.5 times the median maximum amplitude of the Fourier transform of the preceding 50 ms windows) were determined empirically by comparing the automatic artifact rejection outcome to expert assessment of the same data (data assessment performed by HC, KD and LB).</p><p indent="no"><italic>Spike detection</italic>. All signal segments exceeding four times the estimated noise level are flagged as events. Consecutive events are combined into biphasic spike candidates. By comparing spike candidates to a spike template, a spike candidate is either accepted as a spike or is flagged as false positive (figure <figref linkend="jne385116fig02" override="yes">2(B)</figref>). Our spike template imposes a peak-to-peak spike width of <1 ms and total spike duration of <3 ms. No further spike sorting is performed for the present work.</p></sec-level3><sec-level3 id="jne385116s2-2-2" label="2.2.2"><heading>Feature set</heading><p indent="no">For reliable STN detection, robust and distinctive MER features need to be extracted. We computed <italic>noise level</italic> (i), <italic>compound firing rate</italic> (ii) and measures based on PSD (iii), <italic>low band index</italic> (3–12 Hz), <italic>beta band index</italic> (13–30 Hz) and <italic>gamma band index</italic> (31–100 Hz), and determined which features showed the strongest correlation with the surgical annotations. In agreement with previous work, we found that (ordered with decreasing selectivity) noise level, compound firing rate, gamma and beta power were good indicators of the presence of the STN [<cite linkend="jne385116bib12">12</cite>, <cite linkend="jne385116bib13">13</cite>, <cite linkend="jne385116bib15">15</cite>, <cite linkend="jne385116bib20">20</cite>]. Changes in power in the 3–12 Hz range were not specific to the STN. Therefore the low band index has not been included in the feature set used in automatic localization of the STN. On average noise level, compound firing rate and PSD were calculated from MERs of duration 15 ± 6 s.</p><p indent="no"><italic>Noise level</italic>. Noise level is determined by re-applying the envelope-based method to the MER data after initial automatic artifact removal [<cite linkend="jne385116bib23">23</cite>].</p><p indent="no"><italic>Compound firing rate</italic>. Compound firing rate at a specific site is the ratio between the total number of detected spikes and the total recording period in seconds.</p><p indent="no"><italic>Power spectral density</italic>. The PSD of the MER is obtained by computing the power spectrum of the rectified MER after the artifacts have been automatically removed and the mean is subtracted [<cite linkend="jne385116bib15">15</cite>]. The power spectrum of the rectified signal is estimated using Welch's method with a window length of 1 s and 50% overlapping windows. In order to quantify the oscillatory patterns, we defined the low band index (3–12 Hz), beta band index (13–30 Hz) and gamma band index (31–100 Hz). All indices are computed by taking the ratio between the average power in the pre-defined frequency band and the average power in the spectrum.</p></sec-level3><sec-level3 id="jne385116s2-2-3" label="2.2.3"><heading>STN and SNr detection</heading><p indent="no">A threshold-based algorithm is constructed for detection of the STN and the SNr. The algorithm uses as inputs the estimated noise level, the compound firing rate, beta band and gamma band indices observed at each site along a trajectory. Details on noise level, firing rate, beta and gamma band thresholds can be found in the supplementary material (SM2) available at <stacks mmedia="yes"></stacks>.</p><p>Depending on combinations of measures used in delineation of the STN borders, the algorithm allocates different qualitative confidence levels: high, medium or low (figures <figref linkend="jne385116fig03">3</figref> and <figref linkend="jne385116fig04">4</figref>).
<figure id="jne385116fig03" width="page"><graphic><graphic-file version="print" format="EPS" filename="images/jne385116fig03.eps" width="31pc"></graphic-file><graphic-file version="ej" format="JPEG" filename="images/jne385116fig03.jpg"></graphic-file></graphic><caption id="jne385116fc03" label="Figure 3"><p indent="no">Schematic representation of the automatic STN detection algorithm. At each recording site, first, noise level is estimated, artifacts are automatically removed and spikes are detected. PSD is computed, after artifacts have been removed. Then, noise level, firing rate, beta (13–30 Hz) and gamma (31–100 Hz) band indices are computed at each recording site of a trajectory. Per trajectory, noise level, firing rate, beta and gamma band thresholds are calculated. Depending on the combination of features, which exceed the threshold, annotations are made per trajectory using three different confidence levels: high, medium and low.</p></caption></figure><figure id="jne385116fig04" width="page"><graphic><graphic-file version="print" format="EPS" filename="images/jne385116fig04.eps" width="31pc"></graphic-file><graphic-file version="ej" format="JPEG" filename="images/jne385116fig04.jpg"></graphic-file></graphic><caption id="jne385116fc04" label="Figure 4"><p indent="no">(A) Annotation made based on noise level, firing rate and PSD changes. Site <italic>x</italic> is at −2.5 mm, where all measures exceed their respective thresholds. The dorsal border is at −2.5 mm (site <italic>y</italic>) and the ventral border is at 2 mm (site <italic>yy</italic>). For this trajectory, the automatic annotation is in full agreement with the surgical annotations. (B) Region which exceeds the noise level threshold (at −1.5 to −1 mm) does not coincide with the region which exceeds the firing rate threshold (−2 mm). The algorithm has indicated that the STN spans from −2 to −1 mm. The surgical team has annotated the region from −3 to −0.5 mm as the STN. (C) Annotation has been made based on firing rate and PSD. The algorithm has annotated the region from 1 to 2.5 mm as the STN. In this example, the automated annotation fully agrees with the surgical annotation. Beta band (13–30 Hz) and gamma band (31–100 Hz) indices are equivalent to 10log(average power in 13–30 Hz/average power) and 10log(average power in 31–100 Hz/average power), respectively.</p></caption></figure></p><p indent="no"><italic>Annotations with high confidence</italic>. High confidence STN detection is achieved when both increased noise level and increased neuronal unit activity (high firing rate and raised beta or gamma PSD) in a particular MER are detected. From the dorsal to ventral, the first site along the trajectory exceeding the noise level threshold, the firing rate threshold and either the beta band threshold or the gamma band threshold are marked as being within the target structure (site <italic>x</italic>) and the corresponding trajectory is flagged as containing the STN. From site <italic>x</italic> the algorithm searches the dorsal and ventral boundaries by marking the sites (<italic>y</italic> and <italic>yy</italic> respectively) where MER noise level drops and stays below the noise level threshold. The dorsal boundary is further refined by incrementally shifting the boundary in the dorsal direction (site <italic>w</italic>) to include consecutive sites exceeding the firing rate threshold and either the beta band threshold or the gamma band threshold.</p><p indent="no"><italic>Annotations with medium confidence</italic>. In case the firing rate threshold is not exceeded within the region, which has exceeded the noise level threshold, the algorithm searches for consecutive sites of MERs exceeding the noise level threshold, which are subsequently marked as STN. If the sites exceeding the firing rate threshold also exceed either the beta band threshold or the gamma band threshold, precede and are adjacent to the consecutive sites exceeding the noise level threshold, the dorsal border of the STN is shifted to include the sites exceeding the firing rate threshold.</p><p indent="no"><italic>Annotations with low confidence</italic>. In case the noise level threshold is not exceeded in a given trajectory, the algorithm searches for consecutive clusters of MERs exceeding the firing rate threshold and either the beta band threshold or the gamma band threshold. These sites are subsequently marked as STN.</p><p>From the ventral border of the STN (site <italic>yy</italic>), the algorithm searches for a supra-threshold increase in the noise level and the firing rate, and annotates these sites as SNr. The algorithm requires a gap of one site recording between STN and SNr for annotation. For the cases that in the trajectory, there is no gap between STN and SNr, the algorithm is not able to annotate SNr.</p></sec-level3></sec-level2><sec-level2 id="jne385116s2-3" label="2.3"><heading>Statistical analyses used to determine the feature set</heading><p indent="no">Correlation of features (i.e. noise level, compound firing rate and low band, beta band and gamma band indices) with STN anatomy was determined by comparing feature values inside the surgically annotated STN with the ones outside the STN. Only trajectories containing surgically annotated STNs of length greater than 2 depths have been included in the analysis. We applied a paired <italic>t</italic>-test to detect changes in noise level, compound firing rate and low band, beta band and gamma band indices observed on average inside the first half and the second half of surgically annotated STN to average values outside the STN (<italic>p</italic> < 0.01 indicated statistical significance).</p><p>Using surgical annotations to define the dorsal and ventral borders of the STN, we computed if a measure was significantly higher (<italic>p</italic> < 0.05) at the dorsal border of the STN than the average value of a measure observed at the preceding sites and if measures observed at the site following the ventral border of the STN were significantly less (<italic>p</italic> < 0.05) than the average value of the measure observed inside the STN.</p></sec-level2><sec-level2 id="jne385116s2-4" label="2.4"><heading>Validation of the algorithm</heading><p indent="no">Validation is performed by applying the automatic algorithm to the MERs and comparing its output to the surgical (258 trajectories, 6064 sites) or expert annotations (42 trajectories, 1011 sites). The agreement percentage between different annotation sets is computed by calculating the ratio between the number of sites that the two annotations agree on and the total number of sites. Inter-rater reliability is calculated based on Cohen's kappa statistic [<cite linkend="jne385116bib24">24</cite>]. Taking surgical annotations as the golden standard, false negatives were defined as the number of trajectories annotated by the surgical team and not by the algorithm. False positives were defined as the number of trajectories annotated by the algorithm and not by the surgical team. Moreover, we tested the accuracy of the algorithm in detecting the dorsal and the ventral borders and the center of the STN by calculating the difference between the surgically annotated ventral and dorsal STN borders and the automatically detected borders.</p></sec-level2></sec-level1><sec-level1 id="jne385116s3" label="3"><heading>Results</heading><sec-level2 id="jne385116s3-1" label="3.1"><heading>Predictive value of noise level, firing rate and PSD</heading><p indent="no">Based on 239 out of 258 trajectories, where surgical annotations indicated the presence of the STN, we assessed (a) if measures such as noise level, firing rate and PSD are significantly different inside the STN from the values observed outside the STN, and (b) in what percentage of the trajectories these measures were significantly different from the values preceding the surgically annotated dorsal boundary and ventral boundary of the STN.</p><p>Noise level, compound firing rate and beta band and gamma band indices all showed higher values throughout the STN (<italic>p</italic> < 0.01). The low band index showed no differentiation between outside and inside the STN (<italic>p</italic> < 0.01).</p><p>Additionally, using surgical annotations as indicators of the dorsal and the ventral borders of the STN, we computed if a measure was significantly higher (<italic>p</italic> < 0.05) at the dorsal border of the STN than the average value of the measure observed at the preceding sites and if measures observed at the site following the ventral border of the STN were significantly less (<italic>p</italic> < 0.05) than the average value of the measure observed inside of the STN (table <tabref linkend="jne385116tab01">1</tabref>). Results indicate that noise level and firing rate increase are good indicators for the dorsal border of the STN. Moreover, the decrease in noise level and firing rate between inside the STN and the site following the ventral border (exiting the STN) are not as pronounced as the increase in noise level and firing rate between outside the STN and the site marking the dorsal border (entering the STN).
<table id="jne385116tab01" frame="topbot"><caption id="jne385116tc01" label="Table 1"><p indent="no">Percentage of trajectories (<italic>n</italic> = 239) where a measure was significantly higher (<italic>p</italic> < 0.05) at the dorsal border of the STN than the average value of a measure observed at the preceding sites and where a measure observed at the site following the ventral border of the STN was significantly less (<italic>p</italic> < 0.05) than the average value of the measure observed inside the STN.</p></caption><tgroup cols="5"><colspec colnum="1" colname="col1" align="left"></colspec><colspec colnum="2" colname="col2" align="left"></colspec><colspec colnum="3" colname="col3" align="left"></colspec><colspec colnum="4" colname="col4" align="left"></colspec><colspec colnum="5" colname="col5" align="left"></colspec><thead><row><entry></entry><entry>Noise level</entry><entry>Firing rate</entry><entry>Beta band</entry><entry>Gamma band</entry></row></thead><tbody><row><entry>Dorsal</entry><entry>70%</entry><entry>47%</entry><entry>18%</entry><entry>17%</entry></row><row><entry>Ventral</entry><entry>36%</entry><entry>26%</entry><entry>19%</entry><entry>28%</entry></row></tbody></tgroup></table></p></sec-level2><sec-level2 id="jne385116s3-2" label="3.2"><heading>Comparison to expert annotations</heading><sec-level3 id="jne385116s3-2-1" label="3.2.1"><heading>Detection of the STN</heading><p indent="no">The automated algorithm had 88% agreement with surgical annotations made on 84 patient data sets (258 trajectories, 6064 sites) and 87–88% agreement with the two experts’ independent annotations made on 42 randomly chosen trajectories (1011 sites). Additionally, we have compared the agreement level between the clinical experts and each expert with the surgical annotations made on the 42 trajectories in order to establish the inter-rater reliability (table <tabref linkend="jne385116tab02">2</tabref>).
<table id="jne385116tab02" frame="topbot"><caption id="jne385116tc02" label="Table 2"><p indent="no">Average agreement percentages (AAP) and inter-rater reliability (IRR) calculated over 42 randomly chosen trajectories (<italic>n</italic> = 1011): agreement percentage is the ratio between the total number of sites that the two sets of annotations agree on and the total number of sites recorded. Inter-rater reliability is based on Cohen's kappa statistic.</p></caption><tgroup cols="5"><colspec colnum="1" colname="col1" align="left"></colspec><colspec colnum="2" colname="col2" align="left"></colspec><colspec colnum="3" colname="col3" align="left"></colspec><colspec colnum="4" colname="col4" align="left"></colspec><colspec colnum="5" colname="col5" align="left"></colspec><thead><row><entry></entry><entry></entry><entry>Surgical</entry><entry>Expert 1</entry><entry>Expert 2</entry></row><row><entry></entry><entry>Automatic</entry><entry>annotation</entry><entry></entry><entry></entry></row><row><entry></entry><entry>algorithm</entry><entry>(AAP–IRR)</entry><entry>(AAP–IRR)</entry><entry>(AAP–IRR)</entry></row></thead><tbody><row><entry>Automatic</entry><entry></entry><entry>89%–0.75</entry><entry>88%–0.73</entry><entry>87%–0.71</entry></row><row><entry>algorithm</entry><entry></entry><entry></entry><entry></entry><entry></entry></row><row><entry>Surgical</entry><entry></entry><entry></entry><entry>90%–0.77</entry><entry>90%–0.78</entry></row><row><entry>annotation</entry><entry></entry><entry></entry><entry></entry><entry></entry></row><row><entry>Expert 1</entry><entry></entry><entry></entry><entry></entry><entry>93%–0.85</entry></row><row><entry>Expert 2</entry><entry></entry><entry></entry><entry></entry><entry></entry></row></tbody></tgroup></table></p><p>Out of 258 trajectories, the surgical team has indicated the presence of the STN in 239 trajectories and the absence of the STN in 19 trajectories. Taking surgical annotations made on 258 trajectories as the golden standard, our automatic algorithm has made true positive annotations in 231 trajectories and true negative annotations in 12 trajectories. The algorithm has performed false positive annotation of the STN in 7 trajectories and false negative annotation in 8 trajectories (table <tabref linkend="jne385116tab03">3</tabref>). From any given data set, STN annotations have been made on one or more trajectories and false negatives do not occur at any of the channels, which have been later on chosen as the final electrode position by the surgical team.
<table id="jne385116tab03" frame="topbot"><caption id="jne385116tc03" label="Table 3"><p indent="no">Average percentages of agreement with the surgical annotations indicate that annotations made with high confidence have the highest agreement with the surgical annotations. The value of true positive annotations indicates the total number of trajectories containing the STN according to both the surgical team and the automated algorithm. The value of false positive annotations indicates the total number of trajectories containing the STN according to the automated algorithm but not the surgical team. The value of true negative annotations indicates the total number of trajectories not containing the STN according to both the surgical team and the automated algorithm. The value of false negative annotations indicates the total number of trajectories not containing the STN according to the automated algorithm but not the surgical team.</p></caption><tgroup cols="6"><colspec colnum="1" colname="col1" align="left"></colspec><colspec colnum="2" colname="col2" align="left"></colspec><colspec colnum="3" colname="col3" align="left"></colspec><colspec colnum="4" colname="col4" align="left"></colspec><colspec colnum="5" colname="col5" align="left"></colspec><colspec colnum="6" colname="col6" align="left"></colspec><thead><row><entry></entry><entry>Agreement</entry><entry></entry><entry></entry><entry></entry><entry></entry></row><row><entry>Trajectory</entry><entry>with surgical</entry><entry>True</entry><entry></entry><entry>True</entry><entry>False</entry></row><row><entry>annotation</entry><entry>annotations</entry><entry>positive</entry><entry>positive</entry><entry>negative</entry><entry>negative</entry></row></thead><tbody><row><entry>High</entry><entry>88% (<italic>n</italic> = 5192)</entry><entry>216</entry><entry>4</entry><entry></entry><entry></entry></row><row><entry>Medium</entry><entry>75% (<italic>n</italic> = 141)</entry><entry>6</entry><entry>0</entry><entry></entry><entry></entry></row><row><entry>Low</entry><entry>86% (<italic>n</italic> = 262)</entry><entry>9</entry><entry>3</entry><entry></entry><entry></entry></row><row><entry>No STN</entry><entry>94% (<italic>n</italic> = 469)</entry><entry></entry><entry></entry><entry>12</entry><entry>8</entry></row><row><entry>annotated</entry><entry></entry><entry></entry><entry></entry><entry></entry><entry></entry></row><row><entry>Overall</entry><entry>88% (<italic>n</italic> = 6064)</entry><entry>231</entry><entry>7</entry><entry>12</entry><entry>8</entry></row></tbody></tgroup></table></p></sec-level3><sec-level3 id="jne385116s3-2-2" label="3.2.2"><heading>Detection of the dorsal and ventral border of the STN</heading><p indent="no">Over 231 trajectories are identified to contain the STN both by the surgical team and by the automatic algorithm; the accuracy in detecting the dorsal and the ventral border is −0.5 mm/0 mm/0.5 mm (15/50/85 percentiles of border detection errors) (table <tabref linkend="jne385116tab04">4</tabref>). Errors made on detection of the dorsal and the ventral borders of the STN are not symmetric around the respective mean error; dorsal detection errors are spread to the left of the mean error (i.e. ventral) and ventral detection errors are spread to the right of the mean error (i.e. dorsal).
<table id="jne385116tab04" frame="topbot"><caption id="jne385116tc04" label="Table 4"><p indent="no">Algorithm's accuracy in detecting the dorsal and the ventral borders and the center of the STN are computed by taking the difference between the annotations made by the surgical team and the automatic algorithm (231 trajectories). The dorsal border, ventral border and center of the STN detection errors are characterized through the 15/50/85 percentiles and skewness of the detection errors. Positive values indicate that annotations made by the algorithm are more dorsal than the surgical annotations. Negative values indicate that annotations made by the algorithm are more ventral than the surgical annotations. Skewness is a measure for the asymmetry of the detection errors around the sample mean. A positive value for skewness indicates that the detection errors spread to the right of the sample mean (i.e. dorsal); a negative value indicates that the detection errors spread to the left of the sample mean (i.e. ventral). When skewness is equivalent to zero, the data are distributed evenly to the right and left of the mean (e.g. normal distribution).</p></caption><tgroup cols="4"><colspec colnum="1" colname="col1" align="left"></colspec><colspec colnum="2" colname="col2" align="left"></colspec><colspec colnum="3" colname="col3" align="left"></colspec><colspec colnum="4" colname="col4" align="left"></colspec><thead><row><entry></entry><entry>Detection of the</entry><entry>Detection of the</entry><entry>Detection of the</entry></row><row><entry>Algorithm</entry><entry>dorsal border</entry><entry>STN center</entry><entry>ventral border</entry></row><row><entry>confidence</entry><entry>(skewness & 15/50/85</entry><entry>(skewness & 15/50/85</entry><entry>(skewness & 15/50/85</entry></row><row><entry></entry><entry>percentiles)</entry><entry>percentiles)</entry><entry>percentiles)</entry></row></thead><tbody><row><entry>High</entry><entry>−0.7122</entry><entry>0.6658</entry><entry>1.7703</entry></row><row><entry>(5107 sites)</entry><entry>(−0.5/0/0.5 mm)</entry><entry>(−0.5/0/0.5 mm)</entry><entry>(−0.5/0/0.5 mm)</entry></row><row><entry>Medium</entry><entry>1.0977</entry><entry>0.6828</entry><entry>0.1798</entry></row><row><entry>(141 sites)</entry><entry>(−1.8/−0.75/1 mm)</entry><entry>(0.35/0.875/2.45 mm)</entry><entry>(0.2/1.5/3.8 mm)</entry></row><row><entry>Low</entry><entry>−0.5619</entry><entry>0.7778</entry><entry>−0.9419</entry></row><row><entry>(193 sites)</entry><entry>(−1.725/−0.5/0.575 mm)</entry><entry>(−0.5/0/0.825 mm)</entry><entry>(−1.575/0/0.075 mm)</entry></row></tbody></tgroup></table></p></sec-level3></sec-level2></sec-level1><sec-level1 id="jne385116s4" label="4"><heading>Discussion</heading><p indent="no">In this study, we have developed and validated an algorithm for automated detection of the STN from MERs. We tested our method by comparing it to surgical annotations and found a high agreement level (88%) and good inter-rater reliability (0.71–0.73), comparable to the inter-rater reliability between different experienced raters.</p><p>Our novel automated algorithm outlined in this study can be implemented to run in real time to assist clinicians in STN localization during DBS surgery. This would increase the reliability of MER interpretation and reduce surgical time. To test the algorithm's robustness as needed for intra-operative usage, all levels of MER processing have been applied to a large number of consecutive trajectories without data pre-selection, (manual) data pre-processing or data cleaning. We thus aimed to avoid any bias toward clean ‘easier-to-interpret’ recordings and to give an estimate of the reliability of our algorithm in ‘real life’ situations. Statistical analyses demonstrated significant changes in noise level, firing rate and PSD-based measures inside the STN. However, individually, these measures are sub-optimal indicators of the STN boundaries (table <tabref linkend="jne385116tab03">3</tabref>). Therefore our automatic algorithm makes use of combinations of features to detect the STN. The algorithm assigns a qualitative confidence level (high, medium, low) based on the combination of features used in the detection of the STN (figures <figref linkend="jne385116fig03">3</figref> and <figref linkend="jne385116fig04">4</figref>).</p><p>In 92% of the cases (220 out of 238 automatic annotations) the algorithm detected the STN based on typical STN hallmark MER features: increased background noise, increased firing rate, and oscillatory activity, and automatic annotations have been made with high confidence level [<cite linkend="jne385116bib01">1</cite>, <cite linkend="jne385116bib06">6</cite>, <cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15">12–15</cite>]. In these cases, the agreement with surgical annotations (88%) is excellent and comparable to agreement between surgical annotations and off-line expert annotations. In the remaining cases, the three MER measures were not detected to be <italic>simultaneously</italic> increased in any individual recording and hence STN detection was performed with lower level of confidence (medium in 6/238 and low in 12/238). Combining measures for STN detection increases the accuracy of the algorithm and makes it more robust. Substantial changes in the thresholds used in the algorithm have limited effect on the accuracy of the algorithm (supplementary materials available at <stacks mmedia="yes"></stacks>).</p><p>The algorithm has detected the STN in seven trajectories, which have not been annotated by the surgical team (i.e. false positives). Three of these annotations were made with low confidence. Visual inspection of the remaining four cases, where the STN has been detected erroneously, revealed that the noise level estimate has been corrupted by low amplitude artifacts embedded in the noise, giving rise to artificially elevated noise levels. Moreover, eight trajectories, which have been annotated by the surgical team, have been missed by the algorithm (i.e. false negatives). This is due to noise level changes remaining sub-threshold and consecutive sites not exhibiting high firing rate. Decreasing the noise level threshold to reduce the number of false negatives would hamper the accuracy of dorsal and ventral border detection more than increasing the accuracy of detection.</p><p>The accuracy of the algorithm in detecting the ventral and the dorsal borders of the STN is −0.5 mm/0 mm/0.5 mm (15/50/85 percentiles of border detection errors). The estimated cumulative distribution function (cdf) of the dorsal border detection error is skewed in the ventral direction, i.e. the tail of the cdf in the ventral direction is longer than the tail of the cdf in the dorsal direction (table <tabref linkend="jne385116tab04">4</tabref>; supplementary material available at <stacks mmedia="yes"></stacks>), while the cdf of the ventral border detection error is skewed in the dorsal direction, i.e. the tail of the cdf in the dorsal direction is longer than the tail of the cdf in the ventral direction (table <tabref linkend="jne385116tab04">4</tabref>; supplementary material available at <stacks mmedia="yes"></stacks>). The skewness of the cdfs is a direct consequence of the outliers in the dorsal and the ventral border detection errors. Feature thresholds such as the noise level threshold are dependent on the feature values encountered along the trajectory (supplementary material available at <stacks mmedia="yes"></stacks>). Along some trajectories, a transient increase followed by a transient decrease was observed in the noise level as the trajectory passed through the STN. This gave rise to a conservative estimate for where the dorsal and the ventral borders lie, resulting in the ventral skew in the cdf for the dorsal border detection error and the dorsal skew in the cdf for the ventral border detection error. Deviation from the center of the STN and the dorsal and the ventral border detection errors made by the automated algorithm are comparable to the corresponding errors made by the two expert MER raters, which indicates that the annotations made can differ from one expert to another and the variability of the automated algorithm is within the variability of the expert annotations (supplementary material available at <stacks mmedia="yes"></stacks>).</p><p>In the present algorithm, noise level detection is important both for STN boundary detection and for spike detection and artifact rejection. Hence an accurate and reliable estimate of the noise level is required. Several methods have been proposed for noise level estimation either using the root mean square (RMS) of the signal, the median of the distribution of the absolute value of the signal, or the envelope of the signal [<cite linkend="jne385116bib23">23</cite>, <cite linkend="jne385116bib25" range="jne385116bib25,jne385116bib26,jne385116bib27">25–27</cite>]. Previously, it has been demonstrated that the presence of high frequency firing and signal artifacts could result in over-estimation of the noise level for the RMS- and median-based methods, markedly leading to sub-optimal spike detection [<cite linkend="jne385116bib23">23</cite>, <cite linkend="jne385116bib27">27</cite>]. Noise level estimation based on the envelope of the signal is used in this study since it is more robust in the presence of high frequency firing and artifacts [<cite linkend="jne385116bib23">23</cite>]. Moreover, the envelope of the signal can also be used to determine which segments of the recording are artifacts since in this case the presence of high amplitude artifacts hardly influences estimation of the noise level. The use of the envelope of the signal removes an important previous constraint in automatic MER signal processing which is that an artifact cannot be reliably detected without first estimating the noise level and that the noise level cannot be estimated accurately in the presence of artifacts when methods based on the standard deviation or the median of the signal are being used.</p><p>In a recent study by Marceglia <italic>et al</italic>, the importance of standardized methods, which can be applied for the analysis of MERs, has been highlighted. It has been suggested to summarize MERs based on the stability and density of the activity patterns observed [<cite linkend="jne385116bib28">28</cite>]. It is also proposed to evaluate each track based on the features expressed and the quality of these features [<cite linkend="jne385116bib28">28</cite>]. The algorithm described in this study is in line with the proposed method of summarizing MERs. Noise level, compound firing rate and PSD-based measures inherently capture the density of the activity patterns observed while the confidence levels allocated by the algorithm are directly dependent on which features are expressed along each track and how selective the expressed features are for the detection of the STN.</p><p>Several studies have outlined methods for automatic detection and visualization of the STN based on objective and quantitative MER features [<cite linkend="jne385116bib12" range="jne385116bib12,jne385116bib13,jne385116bib14,jne385116bib15,jne385116bib16">12–16</cite>]. In agreement with Wong <italic>et al</italic>, who described a STN visualization algorithm, we have observed that background noise and firing rate most consistently differentiated the STN from its surroundings. Falkenberg <italic>et al</italic> demonstrated that appropriate visualization of power-spectrum-based measures overlapped well with surgical annotations. Zaidel <italic>et al</italic> combined PSD and noise level to locate the STN and its sub-territories. Sub-territories were determined based on changes in the 13–30 Hz frequency band and these changes indicated oscillatory segments of the STN [<cite linkend="jne385116bib15">15</cite>]. In this study, we have investigated changes in the 3–12 Hz (i.e. low band index), 13–30 Hz (i.e. beta band index) and 31–100 Hz (i.e. gamma band index) frequency bands. We have observed that the units inside the STN show significantly higher activities in the beta and gamma bands than the units outside of the STN (<italic>p</italic> < 0.01). In agreement with Zaidel <italic>et al</italic> we observed that these parameters per trajectory are valuable for localizing oscillatory regions within the STN. On the other hand, compared to noise level and compound firing rate, these parameters are less reliable for delineating the dorsal and ventral borders of the STN (table <tabref linkend="jne385116tab01">1</tabref>). Measures based on PSD, though, can be used for STN localization, provided they are combined with other measures such as noise level and firing rate.</p><p>In summary, the algorithm presented in this study reliably detects trajectories containing the STN and identifies the dorsal and the ventral borders of the STN with good accuracy. The automatic results show good consistency with the ratings made by experienced raters. The use of a simple threshold-based method opens the door for this method to be implemented real time during surgery leading to an on-line, easy and objective tool to support STN localization during DBS surgery.</p></sec-level1></body><back><references><heading>References</heading><reference-list type="numeric"><book-ref id="jne385116bib01"><authors><au><second-name>Israel</second-name><first-names>Z</first-names></au><au><second-name>Burchiel</second-name><first-names>K</first-names></au></authors><year>2004</year><book-title>Microelectrode Recording in Movement Disorder Surgery</book-title><publication><place>New York</place><publisher>Thieme Medical Publishers</publisher></publication><pages>pp 63–110</pages></book-ref><journal-ref id="jne385116bib02"><authors><au><second-name>Romanelli</second-name><first-names>P</first-names></au><others><italic>et al</italic></others></authors><year>2004</year><art-title>Microelectrode recording revealing a somatotopic body map in the subthalamic nucleus in humans with Parkinson disease</art-title><jnl-title>J. Neurosurg.</jnl-title><volume>100</volume><pages>611–8</pages><crossref><cr_doi>http://dx.doi.org/10.3171/jns.2004.100.4.0611</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib03"><authors><au><second-name>Priori</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2003</year><art-title>Do intraoperative microrecordings improve subthalamic nucleus targeting in stereotactic neurosurgery for Parkinson's disease?</art-title><jnl-title>J. Neurosurg. Sci.</jnl-title><volume>47</volume><pages>56–60</pages></journal-ref><journal-ref id="jne385116bib04"><authors><au><second-name>Molinuevo</second-name><first-names>J</first-names></au><au><second-name>Valldeoriola</second-name><first-names>F</first-names></au><au><second-name>Valls-Sole</second-name><first-names>J</first-names></au></authors><year>2003</year><art-title>Usefulness of neurophysiologic techniques in stereotactic subthalamic nucleus stimulation for advanced Parkinson's disease</art-title><jnl-title>Clin. Neurophysiol.</jnl-title><volume>114</volume><pages>1793–9</pages><crossref><cr_doi>http://dx.doi.org/10.1016/S1388-2457(03)00160-3</cr_doi></crossref></journal-ref><book-ref id="jne385116bib05"><authors><au><second-name>Kim</second-name><first-names>M</first-names></au><others><italic>et al</italic></others></authors><year>2006</year><art-title>Microelectrode recording: lead point in STN-DBS surgery</art-title><book-title>Advances in Functional and Reparative Neurosurgery <upright>vol 99</upright></book-title><publication><place>Vienna</place><publisher>Springer</publisher></publication><pages>pp 37–42</pages></book-ref><journal-ref id="jne385116bib06"><authors><au><second-name>Hutchinson</second-name><first-names>W</first-names></au><others><italic>et al</italic></others></authors><year>1998</year><art-title>Neurophysiological identification of the subthalamic nucleus in surgery for Parkinson's disease</art-title><jnl-title>Ann. Neurol.</jnl-title><volume>44</volume><pages>622–8</pages><crossref><cr_doi>http://dx.doi.org/10.1002/ana.410440407</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib07"><authors><au><second-name>Benazzouz</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2002</year><art-title>Intraoperative microrecordings of the subthalamic nucleus in Parkinson's disease</art-title><jnl-title>Mov. Disord.</jnl-title><volume>17</volume><pages>S145–9</pages><crossref><cr_doi>http://dx.doi.org/10.1002/mds.10156</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib08"><authors><au><second-name>Hamel</second-name><first-names>W</first-names></au><others><italic>et al</italic></others></authors><year>2003</year><art-title>Subthalamic nucleus stimulation in Parkinson's disease: correlation of active electrode contacts with intraoperative microrecordings</art-title><jnl-title>Stereotact. Funct. Neurosurg.</jnl-title><volume>80</volume><pages>37–42</pages><crossref><cr_doi>http://dx.doi.org/10.1159/000075158</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib09"><authors><au><second-name>Temel</second-name><first-names>Y</first-names></au><others><italic>et al</italic></others></authors><year>2007</year><art-title>Single electrode and multiple electrode guided electrical stimulation of the subthalamic nucleus in advanced Parkinson's disease</art-title><jnl-title>Neurosurgery</jnl-title><volume>61</volume><pages>346–55</pages><crossref><cr_doi>http://dx.doi.org/10.1227/01.neu.0000303993.82149.98</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib10"><authors><au><second-name>Zhuang</second-name><first-names>P</first-names></au><au><second-name>Li</second-name><first-names>Y</first-names></au></authors><year>2003</year><art-title>Characteristics of subthalamic neuronal activities in Parkinson's disease</art-title><jnl-title>Sheng Li Xue Bao</jnl-title><volume>55</volume><pages>435–41</pages></journal-ref><journal-ref id="jne385116bib11"><authors><au><second-name>Starr</second-name><first-names>P</first-names></au></authors><year>2002</year><art-title>Placement of deep brain stimulators into the subthalamic nucleus or <italic>globus pallidus internus</italic>: technical approach</art-title><jnl-title>Stereotact. Funct. Neurosurg.</jnl-title><volume>79</volume><pages>118–45</pages><crossref><cr_doi>http://dx.doi.org/10.1159/000070828</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib12"><authors><au><second-name>Wong</second-name><first-names>S</first-names></au><others><italic>et al</italic></others></authors><year>2009</year><art-title>Functional localization and visualization of the subthalamic nucleus from microelectrode recording acquired during DBS surgery with unsupervised machine learning</art-title><jnl-title>J. Neural Eng.</jnl-title><volume>6</volume><pages>1–11</pages><crossref><cr_doi>http://dx.doi.org/10.1088/1741-2560/6/2/026006</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib13"><authors><au><second-name>Falkenberg</second-name><first-names>J</first-names></au><others><italic>et al</italic></others></authors><year>2006</year><art-title>Automatic analysis and visualization of microelectrode recording trajectories to the subthalamic nucleus: preliminary results</art-title><jnl-title>Stereotact. Funct. Neurosurg.</jnl-title><volume>84</volume><pages>35–45</pages><crossref><cr_doi>http://dx.doi.org/10.1159/000093721</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib14"><authors><au><second-name>Novak</second-name><first-names>P</first-names></au><others><italic>et al</italic></others></authors><year>2007</year><art-title>Detection of the subthalamic nucleus in microelectrographic recordings in Parkinson disease using the high-frequency (>500 Hz) neuronal background</art-title><jnl-title>J. Neurosurg.</jnl-title><volume>106</volume><pages>175–9</pages><crossref><cr_doi>http://dx.doi.org/10.3171/jns.2007.106.1.175</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib15"><authors><au><second-name>Zaidel</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2009</year><art-title>Delimiting subterritories of the human subthalamic nucleus by means of microelectrode recordings and a Hidden Markov Model</art-title><jnl-title>Mov. Disord.</jnl-title><volume>24</volume><pages>1785–93</pages><crossref><cr_doi>http://dx.doi.org/10.1002/mds.22674</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib16"><authors><au><second-name>Moran</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2006</year><art-title>Real time refinement of subthalamic nucleus targeting using Bayesian decision making on the root mean square measure</art-title><jnl-title>Mov. Disord.</jnl-title><volume>21</volume><pages>1425–31</pages><crossref><cr_doi>http://dx.doi.org/10.1002/mds.20995</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib17"><authors><au><second-name>Chen</second-name><first-names>C</first-names></au><others><italic>et al</italic></others></authors><year>2006</year><art-title>Intra-operative recordings of local field potentials can help localize the subthalamic nucleus in Parkinson's disease surgery</art-title><jnl-title>Exp. Neurol.</jnl-title><volume>198</volume><pages>214–21</pages><crossref><cr_doi>http://dx.doi.org/10.1016/j.expneurol.2005.11.019</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib18"><authors><au><second-name>Chen</second-name><first-names>C</first-names></au><others><italic>et al</italic></others></authors><year>2007</year><art-title>Excessive synchronization of basal ganglia neurons at 20 Hz slows movement in Parkinson's disease</art-title><jnl-title>Exp. Neurol.</jnl-title><volume>205</volume><pages>214–21</pages><crossref><cr_doi>http://dx.doi.org/10.1016/j.expneurol.2007.01.027</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib19"><authors><au><second-name>Brown</second-name><first-names>P</first-names></au><au><second-name>Williams</second-name><first-names>D</first-names></au></authors><year>2005</year><art-title>Basal ganglia local field potential activity: character and functional significance in the human</art-title><jnl-title>Clin. Neurophysiol.</jnl-title><volume>116</volume><pages>2510–9</pages><crossref><cr_doi>http://dx.doi.org/10.1016/j.clinph.2005.05.009</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib20"><authors><au><second-name>Trottenberg</second-name><first-names>T</first-names></au><others><italic>et al</italic></others></authors><year>2006</year><art-title>Subthalamic gamma activity in patients with Parkinson's disease</art-title><jnl-title>Exp. Neurol.</jnl-title><volume>200</volume><pages>56–65</pages><crossref><cr_doi>http://dx.doi.org/10.1016/j.expneurol.2006.01.012</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib21"><authors><au><second-name>Pesenti</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2003</year><art-title>The subthalamic nucleus in Parkinson's disease: power spectral density analysis of neural intraoperative signals</art-title><jnl-title>Neurol. Sci.</jnl-title><volume>24</volume><pages>367–74</pages><crossref><cr_doi>http://dx.doi.org/10.1007/s10072-003-0191-2</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib22"><authors><au><second-name>Moran</second-name><first-names>A</first-names></au><others><italic>et al</italic></others></authors><year>2008</year><art-title>Subthalamic nucleus functional organization revealed by parkinsonian neuronal oscillations and synchrony</art-title><jnl-title>Brain</jnl-title><volume>131</volume><pages>3395–409</pages><crossref><cr_doi>http://dx.doi.org/10.1093/brain/awn270</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib23"><authors><au><second-name>Dolan</second-name><first-names>K</first-names></au><others><italic>et al</italic></others></authors><year>2009</year><art-title>Automatic noise-level detection for extra-cellular micro-electrode recordings</art-title><jnl-title>Med. Biol. Eng. Comput.</jnl-title><volume>47</volume><pages>791–800</pages><crossref><cr_doi>http://dx.doi.org/10.1007/s11517-009-0494-4</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib24"><authors><au><second-name>Cohen</second-name><first-names>J</first-names></au></authors><year>1960</year><art-title>A coefficient of agreement of nominal scales</art-title><jnl-title>Educ. Psychol. Meas.</jnl-title><volume>20</volume><pages>37–46</pages><crossref><cr_doi>http://dx.doi.org/10.1177/001316446002000104</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib25"><authors><au><second-name>Pancrazio</second-name><first-names>J</first-names></au><others><italic>et al</italic></others></authors><year>2003</year><art-title>A portable microelectrode array recording system incorporating cultured neuronal networks for neurotoxin detection</art-title><jnl-title>Biosens. Bioelectron.</jnl-title><volume>18</volume><pages>1339–47</pages><crossref><cr_doi>http://dx.doi.org/10.1016/S0956-5663(03)00092-7</cr_doi></crossref></journal-ref><conf-ref id="jne385116bib26"><authors><au><second-name>Watkins</second-name><first-names>P</first-names></au><others><italic>et al</italic></others></authors><year>2004</year><art-title>Validation of adaptive threshold spike detector for neural recording</art-title><conf-title>Conf. Proc. IEEE Engineering in Medicine and Biology Society (San Francisco, CA)</conf-title><pages>pp 4079–82</pages></conf-ref><journal-ref id="jne385116bib27"><authors><au><second-name>Quiroga</second-name><first-names>R</first-names></au><au><second-name>Nadasdy</second-name><first-names>Z</first-names></au><au><second-name>Ben-Shaul</second-name><first-names>Y</first-names></au></authors><year>2004</year><art-title>Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering</art-title><jnl-title>Neural. Comput.</jnl-title><volume>16</volume><pages>1661–87</pages><crossref><cr_doi>http://dx.doi.org/10.1162/089976604774201631</cr_doi></crossref></journal-ref><journal-ref id="jne385116bib28"><authors><au><second-name>Marceglia</second-name><first-names>S</first-names></au><others><italic>et al</italic></others></authors><year>2010</year><art-title>Multicenter study report: electrophysiological monitoring procedures for subthalamic deep brain stimulation surgery in Parkinson's disease</art-title><jnl-title>Neurol. Sci.</jnl-title><volume>31</volume><pages>449–57</pages><crossref><cr_doi>http://dx.doi.org/10.1007/s10072-010-0254-0</cr_doi></crossref></journal-ref></reference-list></references></back></article></istex:document></istex:metadataXml><mods version="3.6"><titleInfo><title>Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title></titleInfo><titleInfo type="abbreviated"><title>Automatic subthalamic nucleus detection</title></titleInfo><titleInfo type="alternative"><title>Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity</title></titleInfo><name type="personal"><namePart type="given">Hayriye</namePart><namePart type="family">Cagnan</namePart><affiliation>Department of Clinical Neurology, University of Oxford, UK</affiliation><affiliation>Author to whom any correspondence should be addressed</affiliation><affiliation>E-mail:hayriye.cagnan@clneuro.ox.ac.uk</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Kevin</namePart><namePart type="family">Dolan</namePart><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Xuan</namePart><namePart type="family">He</namePart><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Maria Fiorella</namePart><namePart type="family">Contarino</namePart><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Richard</namePart><namePart type="family">Schuurman</namePart><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Pepijn</namePart><namePart type="family">van den Munckhof</namePart><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Wytse J</namePart><namePart type="family">Wadman</namePart><affiliation>SILS, Universiteit van Amsterdam, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Lo</namePart><namePart type="family">Bour</namePart><affiliation>Academic Medical Center, Universiteit van Amsterdam, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><name type="personal"><namePart type="given">Hubert C F</namePart><namePart type="family">Martens</namePart><affiliation>Philips Research Laboratories, Eindhoven, 5656 AE, The Netherlands</affiliation><role><roleTerm type="text">author</roleTerm></role></name><typeOfResource>text</typeOfResource><genre type="research-article" displayLabel="paper"></genre><originInfo><publisher>IOP Publishing</publisher><dateIssued encoding="w3cdtf">2011</dateIssued><copyrightDate encoding="w3cdtf">2011</copyrightDate></originInfo><language><languageTerm type="code" authority="iso639-2b">eng</languageTerm><languageTerm type="code" authority="rfc3066">en</languageTerm></language><physicalDescription><internetMediaType>text/html</internetMediaType><note type="production">Printed in the UK</note></physicalDescription><abstract>Microelectrode recording (MER) along surgical trajectories is commonly applied for refinement of the target location during deep brain stimulation (DBS) surgery. In this study, we utilize automatically detected MER features in order to locate the subthalamic nucleus (STN) employing an unsupervised algorithm. The automated algorithm makes use of background noise level, compound firing rate and power spectral density along the trajectory and applies a threshold-based method to detect the dorsal and the ventral borders of the STN. Depending on the combination of measures used for detection of the borders, the algorithm allocates confidence levels for the annotation made (i.e. high, medium and low). The algorithm has been applied to 258 trajectories obtained from 84 STN DBS implantations. MERs used in this study have not been pre-selected or pre-processed and include all the viable measurements made. Out of 258 trajectories, 239 trajectories were annotated by the surgical team as containing the STN versus 238 trajectories by the automated algorithm. The agreement level between the automatic annotations and the surgical annotations is 88. Taking the surgical annotations as the golden standard, across all trajectories, the algorithm made true positive annotations in 231 trajectories, true negative annotations in 12 trajectories, false positive annotations in 7 trajectories and false negative annotations in 8 trajectories. We conclude that our algorithm is accurate and reliable in automatically identifying the STN and locating the dorsal and ventral borders of the nucleus, and in a near future could be implemented for on-line intra-operative use.</abstract><subject><genre>Keywords</genre><topic></topic></subject></mods></metadata><serie></serie></istex></record>Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Wicri/Sante/explor/ParkinsonV1/Data/Main/Corpus
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000648 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/Main/Corpus/biblio.hfd -nk 000648 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien
|wiki= Wicri/Sante
|area= ParkinsonV1
|flux= Main
|étape= Corpus
|type= RBID
|clé= ISTEX:AE48434E0788CA8C26BA50FD02CDF77DA015C6C1
|texte= Automatic subthalamic nucleus detection from microelectrode recordings based on noise level and neuronal activity
}}
| This area was generated with Dilib version V0.6.23. |  |