ONYIA, KASIEMOBI JESSE (2026) DEVELOPMENT OF AN EXPLAINABLE DEEP LEARNING MODEL FOR MRI-BASED STROKE DETECTION AND LESION SEGMENTATION. Other thesis, Godfrey Okoye University, Enugu Nigeria.
![[img]](https://eprints.gouni.edu.ng/style/images/fileicons/text.png) |
Text
Onyia Kasiemobi Jesse full project.pdf Download (2MB) |
Abstract
This thesis introduces StrokeVision AI, an explainable AI tool to facilitate early identification of ischemic stroke lesion on DWI MR images. The need for this type of tool arose due to limitations in the diagnosis of acute stroke especially for those who present in resource-limited settings and a need for speedy diagnosis with improved accuracy and transparency. StrokeVision AI comprises a 2D U-Net model trained using theISLES2022[98] dataset and a visual explanation of predictions based on Occlusion Sensitivity Mapping(OSM). The complete StrokeVision AI system is designed as an open-source web application using FastAPI, React, MySQL and Docker. Training was performed without evidence of overfitting and exhibited stability. The DSC of0.6751obtained outperformed the generally accepted range of performance for comparable 2D U-Net models. The free access and the ease of deployment of the StrokeVision AI system makes it an extremely relevant tool for the healthcare system in Nigeria and more broadly across sub-Saharan African countries, where resource constraints for diagnosis of stroke are prominent.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Computing And Information Technology (FACIT) |
| Depositing User: | Cynthia Ugwuoti |
| Date Deposited: | 28 Jul 2026 12:45 |
| Last Modified: | 28 Jul 2026 12:45 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6061 |
Actions (login required)
 |
View Item |
