11/19/2025
By Vitaly Dobromyslin

The Francis College of Engineering, Department of Biomedical Engineering, invites you to attend a Doctoral Dissertation Proposal defense by Vitaly Dobromyslin titled: "Integrating fMRI-Derived Biomarkers and Machine Learning for Enhanced Chronic Stroke Infarct Detection."

Date: Wednesday, Dec. 3, 2025
Time: 2 to 3 p.m.
Location: Virtual defense via Zoom. Please contact vitaly_dobromyslin@student.uml.edu for Zoom link

Committee Members:

  • Wenjin Zhou (Advisor), Assistant Professor, Miner School of Computer & Information Sciences, UMass Lowell
  • Benuyan Liu (Member), Professor, Director, Miner School of Computer & Information Sciences, UMass Center for Digital Health (CDH), Computer Networking Lab, CHORDS, UMass Lowell
  • Yu Cao (Member), Professor, Director, Miner School of Computer & Information Sciences, UMass Center for Digital Health (CDH), UMass Lowell

Abstract:
Stroke remains the fifth leading cause of death in the United States and a major source of long-term disability. Some acute (<24h post onset) and sub-acute (1-3 weeks post onset) stroke episodes may go unnoticed by a patient and a healthcare provider, yet they can cause permanent infarct tissue in the oxygen deprived areas of the brain, known as chronic (>3 weeks post onset) infarct. The repeat infarcts have been associated with higher risk of successive infarcts and increased morbidity. Improving early, accurate, and scalable detection is therefore essential for patient outcomes and healthcare efficiency. Existing imaging approaches for infarct detection - such as CT, T1-weighted (T1w) MRI, diffusion-weighted imaging, fluid-attenuated inversion recovery, and contrast-based perfusion—each have limitations for broad screening or longitudinal monitoring. Lower sensitivity and risk from ionizing radiation make CT less preferable than other imaging modalities. While contrast-based perfusion provides high sensitivity to infarcts, it carries health risks due to long term contrast agent deposition in the brain. T1w MRI is widely available and non-invasive, but it provides limited contrast for infarcted tissue and requires substantial interpretation time from radiologists, which constrains throughput and increases costs. Machine learning (ML) techniques, such as deep neural networks, have recently made significant progress in coming close to matching radiologist level performance in detecting certain types of medical anomalies. Previous attempts to apply ML techniques for chronic infarct detection with T1w MRI saw limited clinical adoption partly due: 1) focus on the acute stroke phase during medical intervention, 2) varying detection performance based on lesion size, and 3) lack of evidence linking infarct manifestation to patient prognosis.

In ongoing experiments, we found that resting-state functional MRI (rs-fMRI), tracking blood oxygen level dependent (BOLD) signal, can provide diagnostic information for chronic stroke detection. Our preliminary data show that certain chronic infarct biomarkers can be automatically extracted from the rs-fMRI signal with ML techniques. These data suggest that neuronal viability may be closely related to information theory metrics, such as signal envelope and statistical properties of the residual noise. Given these findings, the hypothesis is that new automated machine learning (auto-ML) BOLD biomarkers can complement state-of-the-art T1-based approaches (e.g., transformers, convolutional neural networks, etc.) for chronic stroke detection and will improve the overall detection performance.