Ahmad Sabri, M.I. and Muda, A.S. and Tharek, A. and Faye, I. and Shamsuddin Perisamy, Rajeev and Abu Bakar, I.S. and Mahfar, N. (2025) AI-assisted detection of Hyperintense Vessel Sign on FLAIR MRI: a novel triage tool for acute ischemic stroke management. In: Malaysian Society of Interventional Radiology (MYSIR) Annual Scientific Meeting 2025, 31st December 2025, Puncak Alam, Selangor.
|
PDF (Abstract)
- Published Version
Download (376kB) | Preview |
Abstract
Introduction: The Hyperintense Vessel Sign (HVS) on FLAIR MRI is a subtle yet critical marker of arterial occlusion in acute ischemic stroke. Its timely detection can influence decisions regarding thrombolysis or thrombectomy eligibility. However, manual HVS identification is time-intensive and prone to inter-observer variability, especially in high-pressure emergency settings. We present a novel deep learning-based triage tool designed to assist radiologists by automating HVS detection with high computational efficiency and clinical reliability. Method: A total of 300 FLAIR MRI datasets were retrospectively collected from Hospital Sultan Abdul Aziz Shah (HSAAS), UPM, obtained using a standardized protocol on a 3T scanner. A deep learning model based on the nnU-Net architecture was developed to detect HVS with pixel-level precision. The model was trained using 5-fold cross-validation and tested against annotations by three board-certified neuroradiologists (gold standard). Inference was conducted on an RTX 4080 GPU with an average runtime of 30 seconds per scan. Novel features included the integration of explainable AI (XAI) techniques to enhance model transparency and improve radiologist trust in AI outputs. Results: The model achieved a sensitivity of 89%, specificity of 84%, and Dice score of 0.78 ± 0.11 compared to radiologists’ consensus annotations (accuracy: 95%). While radiologists outperformed the model diagnostically, the tool reduced average triage decision time by 40%, prioritizing high-risk cases for review without compromising safety. Importantly, XAI visualizations provided interpretable heatmaps highlighting regions of interest, which radiologists reported as valuable for cross-verification during time-critical scenarios. Conclusion: By reducing decision-making time while maintaining diagnostic accuracy, this approach has the potential to transform stroke workflows in resource-limited or high-volume settings. Future work will focus on integrating this tool into real-time clinical pipelines and expanding its application to multi-modal imaging data for comprehensive stroke assessment.
| Item Type: | Proceeding Paper (Plenary Papers) |
|---|---|
| Additional Information: | 7722/127064 |
| Subjects: | Q Science > Q Science (General) R Medicine > R Medicine (General) |
| Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button): | Kulliyyah of Medicine Kulliyyah of Medicine > Department of Radiology |
| Depositing User: | Dr Rajeev Shamsuddin Perisamy |
| Date Deposited: | 29 Jan 2026 09:36 |
| Last Modified: | 01 Feb 2026 10:19 |
| Queue Number: | 2026-01-Q1830 |
| URI: | http://irep.iium.edu.my/id/eprint/127064 |
Actions (login required)
 |
View Item |
Download Statistics
Download Statistics