Related Experiment Video
Updated: Jan 12, 2026
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
49.3K
A multimodal multitask deep learning model for predicting stroke lesion and functional outcomes using 4D CTP imaging
Kimberly Amador1,2, Anthony J Winder3, Jens Fiehler4
1Department of Radiology, University of Calgary, Calgary, Canada. kimberlyalejandra.am@ucalgary.ca.
Scientific Reports
|November 1, 2025
Summary
This study introduces CTPredict, a novel AI model for predicting acute ischemic stroke outcomes. It uses multimodal multitask learning to forecast both lesion progression and functional recovery, improving patient care.
Area of Science:
- Neurology
- Artificial Intelligence
- Medical Imaging
Background:
- Acute ischemic stroke poses significant disability risks.
- Four-dimensional computed tomography perfusion (4D CTP) aids in acute stroke assessment.
- Predicting stroke lesion and functional outcomes is crucial but often done separately.
Purpose of the Study:
- To develop a multimodal, multitask deep learning model for simultaneous prediction of stroke lesion and functional outcomes.
- To leverage shared patterns between lesion and functional outcomes for enhanced prediction accuracy.
- To integrate 4D CTP imaging and clinical metadata for comprehensive stroke outcome prediction.
Main Methods:
- Developed CTPredict, a multimodal, multitask deep learning framework.
- Integrated modality-specific encoders and a multimodal fusion module with cross-attention.
- Employed task-specific branches for predicting follow-up lesions and 90-day modified Rankin Scale scores.
Main Results:
- CTPredict achieved a 0.23 Dice score for lesion prediction and 0.77 accuracy for functional outcome prediction.
- Outperformed single-task models, demonstrating the benefit of multitask learning.
- Validated on a multi-center dataset of 111 acute ischemic stroke patients.
Conclusions:
- Multitask learning enhances the prediction of stroke outcomes by utilizing shared data patterns.
- CTPredict offers a streamlined, data-driven approach for personalized stroke outcome prediction.
- This model shows potential for improving clinical decision-making in acute ischemic stroke management.
