Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models
Stroke has become a leading cause of disability worldwide. Early medication and rehabilitation is the key to help post-stroke survivors recover faster. Presently, doctors rely on imaging modalities like CT/MRI for diagnosing stroke patients. The diagnosis done using these modalities can be highly su...
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-06-01
|
| Series: | Biomedical Engineering Advances |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2667099224000100 |
| _version_ | 1827225781093269504 |
|---|---|
| author | Shatakshi Singh Dimple Dawar Esha Mehmood Jeyaraj Durai Pandian Rajeshwar Sahonta Subhash Singla Amit Batra Cheruvu Siva Kumar Manjunatha Mahadevappa |
| author_facet | Shatakshi Singh Dimple Dawar Esha Mehmood Jeyaraj Durai Pandian Rajeshwar Sahonta Subhash Singla Amit Batra Cheruvu Siva Kumar Manjunatha Mahadevappa |
| author_sort | Shatakshi Singh |
| collection | DOAJ |
| description | Stroke has become a leading cause of disability worldwide. Early medication and rehabilitation is the key to help post-stroke survivors recover faster. Presently, doctors rely on imaging modalities like CT/MRI for diagnosing stroke patients. The diagnosis done using these modalities can be highly subjective. Apart from this, these imaging modalities are very costly, time taking and inconvenient for the patients. So there is a need of faster, portable and an automated diagnostic system for assessing post-stroke conditions so that right measures can be taken in the right time. To cater to this need EEG comes in handy because of its portable nature. So, in this work, utility of EEG has been studied to diagnose three aspects of stroke: 1) type of stoke, 2) affected artery and 3) severity of stroke. To achieve this, one-minute resting state EEG data was used to extract 57 features. The features were ranked and selected using ranking algorithm and deep learning (DL) models were trained with supervision from information extracted using MRI data. To find out type of stroke and affected artery DWI, SWI and MRA images were used, and severity of stroke was recorded in terms of NIHSS score. Three different DL models were trained for each task i.e. type of stroke, affected artery and severity of stroke. For classifying type of stroke an accuracy of 97.74% was obtained using 37 features. For stroke severity, the model gave RMSE of 2.1955 with a high correlation value (r = 0.91). The DL model for classifying affected artery used 33 features and gave accuracy of 95.7%. It was also found that less complex time domain features and QEEG features were frequently selected out of 57 features for all the DL models. Features in delta and theta sub-bands were frequently selected along with QEEG features. The work presented here established that EEG can act as a reliable modality for faster diagnosis of stroke specifics and hence can help medical professionals in speeding the decision making process. |
| first_indexed | 2024-04-24T11:37:04Z |
| format | Article |
| id | doaj.art-33a977d029eb4dfe9b759cf93d6bc9bb |
| institution | Directory Open Access Journal |
| issn | 2667-0992 |
| language | English |
| last_indexed | 2025-03-21T17:31:37Z |
| publishDate | 2024-06-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Biomedical Engineering Advances |
| spelling | doaj.art-33a977d029eb4dfe9b759cf93d6bc9bb2024-06-13T04:46:11ZengElsevierBiomedical Engineering Advances2667-09922024-06-017100121Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning ModelsShatakshi Singh0Dimple Dawar1Esha Mehmood2Jeyaraj Durai Pandian3Rajeshwar Sahonta4Subhash Singla5Amit Batra6Cheruvu Siva Kumar7Manjunatha Mahadevappa8School of Medical Science and Technology, IIT Kharagpur, IndiaDepartment of Neurology, Christian Medical College, Ludhiana, IndiaDepartment of Neurology, Christian Medical College, Ludhiana, IndiaDepartment of Neurology, Christian Medical College, Ludhiana, IndiaDepartment of Neurology, Christian Medical College, Ludhiana, IndiaDepartment of Radiodiagnosis, Christian Medical College, Ludhiana, IndiaDepartment of Radiology, Christian Medical College, Ludhiana, IndiaMechanical Engineering Department, IIT Kharagpur, IndiaSchool of Medical Science and Technology, IIT Kharagpur, India; Corresponding Author: Manjunatha Mahadevappa, PhD, School of Medical Science and Technology Life Science Building, IIT Kharagpur, India.Stroke has become a leading cause of disability worldwide. Early medication and rehabilitation is the key to help post-stroke survivors recover faster. Presently, doctors rely on imaging modalities like CT/MRI for diagnosing stroke patients. The diagnosis done using these modalities can be highly subjective. Apart from this, these imaging modalities are very costly, time taking and inconvenient for the patients. So there is a need of faster, portable and an automated diagnostic system for assessing post-stroke conditions so that right measures can be taken in the right time. To cater to this need EEG comes in handy because of its portable nature. So, in this work, utility of EEG has been studied to diagnose three aspects of stroke: 1) type of stoke, 2) affected artery and 3) severity of stroke. To achieve this, one-minute resting state EEG data was used to extract 57 features. The features were ranked and selected using ranking algorithm and deep learning (DL) models were trained with supervision from information extracted using MRI data. To find out type of stroke and affected artery DWI, SWI and MRA images were used, and severity of stroke was recorded in terms of NIHSS score. Three different DL models were trained for each task i.e. type of stroke, affected artery and severity of stroke. For classifying type of stroke an accuracy of 97.74% was obtained using 37 features. For stroke severity, the model gave RMSE of 2.1955 with a high correlation value (r = 0.91). The DL model for classifying affected artery used 33 features and gave accuracy of 95.7%. It was also found that less complex time domain features and QEEG features were frequently selected out of 57 features for all the DL models. Features in delta and theta sub-bands were frequently selected along with QEEG features. The work presented here established that EEG can act as a reliable modality for faster diagnosis of stroke specifics and hence can help medical professionals in speeding the decision making process.http://www.sciencedirect.com/science/article/pii/S2667099224000100EEGMRIStrokeDeep LearningQEEGNIHSS |
| spellingShingle | Shatakshi Singh Dimple Dawar Esha Mehmood Jeyaraj Durai Pandian Rajeshwar Sahonta Subhash Singla Amit Batra Cheruvu Siva Kumar Manjunatha Mahadevappa Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models Biomedical Engineering Advances EEG MRI Stroke Deep Learning QEEG NIHSS |
| title | Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models |
| title_full | Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models |
| title_fullStr | Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models |
| title_full_unstemmed | Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models |
| title_short | Determining Diagnostic Utility of EEG for Assessing Stroke Severity using Deep Learning Models |
| title_sort | determining diagnostic utility of eeg for assessing stroke severity using deep learning models |
| topic | EEG MRI Stroke Deep Learning QEEG NIHSS |
| url | http://www.sciencedirect.com/science/article/pii/S2667099224000100 |
| work_keys_str_mv | AT shatakshisingh determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT dimpledawar determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT eshamehmood determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT jeyarajduraipandian determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT rajeshwarsahonta determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT subhashsingla determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT amitbatra determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT cheruvusivakumar determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels AT manjunathamahadevappa determiningdiagnosticutilityofeegforassessingstrokeseverityusingdeeplearningmodels |