Related Experiment Video
Updated: Aug 27, 2025
10:33
Advanced Diffusion Imaging in The Hippocampus of Rats with Mild Traumatic Brain Injury
Published on: August 14, 2019
8.6K
Interpretable deep learning for the prognosis of long-term functional outcome post-stroke using acute diffusion
Eric Moulton1, Romain Valabregue1,2, Michel Piotin3
1Institut du Cerveau et de la Moelle épinière, ICM, Inserm U 1127, CNRS UMR 7225, Sorbonne Université, Paris, France.
Summary
Deep learning models using convolutional neural networks (CNNs) show promise in predicting long-term stroke outcomes from diffusion-weighted imaging. These AI models outperform traditional methods, offering better accuracy and interpretability for acute stroke patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Traditional biomarkers for acute stroke functional outcome prediction have limitations.
- Deep learning offers potential for more powerful and explainable prediction models in stroke imaging.
Purpose of the Study:
- To evaluate a deep learning model (CNN) for predicting long-term functional outcome in acute ischemic stroke patients.
- To compare the CNN's performance and interpretability against traditional biomarkers (lesion volume, ASPECTS).
Main Methods:
- A convolutional neural network (CNN) with an attention mechanism was trained on diffusion-weighted imaging (DWI) from 322 ischemic stroke patients.
- The CNN predicted 3-month functional outcome (modified Rankin Scale) and was compared to logistic regression models using lesion volume and ASPECTS.
- The attention mechanism visualized brain areas driving the CNN's predictions.
Main Results:
- The CNN achieved a significantly higher area under the curve (0.83) compared to lesion volume (0.78) and ASPECTS (0.77).
- At equal specificity, the CNN demonstrated significantly higher sensitivity (0.67) than lesion volume (0.48) and ASPECTS (0.50).
- The attention mechanism confirmed the CNN focused on the lesion area for outcome prediction.
Conclusions:
- Deep learning models, specifically CNNs, can accurately predict long-term functional outcomes in acute ischemic stroke using DWI.
- CNNs offer superior performance and interpretability over traditional biomarkers for stroke outcome prediction.
- AI-driven imaging analysis holds significant potential to enhance acute stroke management and patient care.
