Advancing Perioperative Stroke Prevention: Machine Learning-Based Electroencephalography Monitoring for Detecting Cerebral Ischemia During Carotid Endarterectomy

Mina, Amir I. (2024) Advancing Perioperative Stroke Prevention: Machine Learning-Based Electroencephalography Monitoring for Detecting Cerebral Ischemia During Carotid Endarterectomy. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

	PDF Primary Text Restricted to University of Pittsburgh users only until 14 October 2025. Download (3MB) | Request a Copy
	HTML (Interactive plots of feature and label distributions) Supplemental Material Restricted to University of Pittsburgh users only until 14 October 2025. Download (286MB) | Request a Copy
	Other (Anaconda environment) Supplemental Material Restricted to University of Pittsburgh users only until 14 October 2025. Download (9kB) | Request a Copy

Abstract

This research investigates machine learning (ML) to improve the detection of cerebral ischemia during carotid endarterectomy (CEA) procedures, which carry a higher stroke risk. This work aims to advance perioperative stroke prevention, a severe complication posing difficulty in timely diagnosis. Studies show over 80% of patients experiencing perioperative stroke suffer from significant disability or death. Intraoperative neuromonitoring (IONM) can be done using electroencephalography (EEG) to detect ischemic brain changes that may indicate stroke onset. Manual EEG monitoring poses challenges due to subjective interpretation, potential errors, and limited expert availability.
This work explores integrating ML with EEG to overcome these limitations, pioneering the use of intraoperative EEG for real-time ischemia detection. This contrasts with previous studies that used ML to identify post-stroke EEG patterns. The thesis details the development of ML models to detect ischemic patterns using quantitative EEG (qEEG) features and evaluates them through traditional and time series-aware metrics. Their performance is validated against retrospective labels from a panel of five neurophysiologists.
The results demonstrate that tree-based models effectively distinguish ischemic changes using qEEG. The models’ feature importance ranking reflects the clinical understanding of how ischemia presents in EEG. Time series-aware metrics reveal variations in the models’ abilities to predict the timing and duration of ischemia, essential factors for real-time surgical contexts. Importantly, the ML models achieved non-inferior performance compared to the neurophysiologists’ retrospective review. Of note, moderate agreement between the retrospectively refined labels of five neurophysiologists and minimal agreement with the initial intraoperative labels was found. This emphasizes the inherent subjectivity in EEG interpretation and raises concerns about the reliability of manual IONM.
The findings validate the efficacy of these ML models for intraoperative ischemia detection and their potential for acceptance due to clinically sound explainability. This research contributes to the field by laying a foundation for innovation in perioperative stroke prevention by merging ML approaches with established neuromonitoring techniques to improve surgical outcomes and patient safety.

---

Share

Citation/Export:	Select format... Citation - Text Citation - HTML Endnote BibTex Dublin Core OpenURL MARC (ISO 2709) METS MODS EP3 XML Reference Manager Refer
Social Networking:	Share |

---

Details

Item Type:	University of Pittsburgh ETD
Status:	Unpublished
Creators/Authors:	Creators Email Pitt Username ORCID Mina, Amir I. amm500@pitt.edu amm500 0000-0002-1231-1696
ETD Committee:	Title Member Email Address Pitt Username ORCID Committee Chair Hochheiser, Harry harryh@pitt.edu Committee Member Batmanghelich, Kayhan batman@bu.edu Committee Member Thirumala, Parthasarathy thirumalapd@upmc.edu Committee Member Senathirajah, Yalini yalini@pitt.edu Thesis Advisor Visweswaran, Shyam shv3@pitt.edu
Date:	14 October 2024
Date Type:	Publication
Defense Date:	30 July 2024
Approval Date:	14 October 2024
Submission Date:	14 August 2024
Access Restriction:	1 year -- Restrict access to University of Pittsburgh for a period of 1 year.
Number of Pages:	147
Institution:	University of Pittsburgh
Schools and Programs:	School of Medicine > Biomedical Informatics
Degree:	PhD - Doctor of Philosophy
Thesis Type:	Doctoral Dissertation
Refereed:	Yes
Uncontrolled Keywords:	Intraoperative EEG, Real-time Ischemia Detection, Machine Learning Models, Tree-based Models, Quantitative EEG Features, Neurophysiology, Time Series Analysis, Surgical Outcomes, EEG Signal Processing, Intraoperative Stroke Prevention, Ischemic Pattern Recognition, Non-inferiority
Date Deposited:	14 Oct 2024 16:07
Last Modified:	14 Oct 2024 16:07
URI:	http://d-scholarship.pitt.edu/id/eprint/46936

Available Versions of this Item

• Advancing Perioperative Stroke Prevention: Machine Learning-Based Electroencephalography Monitoring for Detecting Cerebral Ischemia During Carotid Endarterectomy. (deposited UNSPECIFIED)
  • Advancing Perioperative Stroke Prevention: Machine Learning-Based Electroencephalography Monitoring for Detecting Cerebral Ischemia During Carotid Endarterectomy. (deposited 14 Oct 2024 16:07) [Currently Displayed]

---

Metrics

Monthly Views for the past 3 years

Plum Analytics

---

Actions (login required)

View Item