Related Experiment Video
Updated: May 29, 2025
13:10
Combined Near-infrared Fluorescent Imaging and Micro-computed Tomography for Directly Visualizing Cerebral Thromboemboli
Published on: September 25, 2016
9.9K
Automatic etiological classification of stroke thrombus digital photographs using a deep learning model
Álvaro Lucero-Garófano1,2, Alicia Aliena-Valero1, Isabel Vielba-Gómez1,3
1Unidad Mixta de Investigación Cerebrovascular, Instituto de Investigación Sanitaria La Fe, Valencia, Spain.
Frontiers in Neurology
|February 3, 2025
Summary
Deep learning models can now automatically classify ischemic stroke causes using thrombus images. This AI approach aids in determining stroke etiology for better secondary prevention strategies.
Area of Science:
- Neurology
- Artificial Intelligence
- Medical Imaging
Background:
- Accurate etiological classification of ischemic stroke is crucial for effective secondary prevention.
- A significant proportion of ischemic strokes have undetermined causes, posing challenges for treatment.
- Mechanical thrombectomy provides retrieved thrombus samples suitable for analysis.
Purpose of the Study:
- To develop and validate a Deep Learning (DL) model for the automatic etiological classification of ischemic stroke.
- To utilize digital images of thrombi obtained during mechanical thrombectomy as input for the DL model.
- To integrate clinical characteristics with imaging data for improved classification accuracy.
Main Methods:
- A DL model comprising two deep neural networks was designed for thrombus image segmentation and etiological classification.
- The model was trained and evaluated on digital thrombus images and clinical data from 166 patients with large vessel occlusion stroke.
- Performance was assessed using DICE coefficient for segmentation and accuracy, precision, sensitivity, specificity, and AUC for classification.
Main Results:
- The segmentation network achieved a high average DICE coefficient of 0.96.
- The fused imaging and clinical classification network demonstrated excellent performance with 0.968 accuracy and 0.947 AUC.
- The model effectively classified cryptogenic thrombi, assigning 96% to cardioembolic and 4% to atherothrombotic etiology.
Conclusions:
- Two convolutional neural networks, combined with clinical data, can accurately segment thrombus images and classify ischemic stroke etiology.
- This DL-based approach offers a precise method for differentiating between atherothrombotic and cardioembolic causes of acute ischemic stroke.
- The model shows promise for improving the etiological diagnosis of ischemic stroke, potentially enhancing secondary prevention strategies.
