Related Experiment Video
Updated: Jul 23, 2025
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
1.1K
Distinct brain morphometry patterns revealed by deep learning improve prediction of aphasia severity
Alex Teghipco1, Roger Newman-Norlund1, Julius Fridriksson1
1University of South Carolina.
Research Square
|July 18, 2023
Summary
Deep learning models using Convolutional Neural Networks (CNNs) accurately predict post-stroke aphasia severity by analyzing brain atrophy patterns beyond the lesion. These models outperform traditional methods, offering improved prognostication for stroke patients.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Imaging
Background:
- Post-stroke aphasia severity is influenced by factors beyond the primary lesion.
- Interindividual variability in aphasia suggests other brain integrity factors are involved.
- Brain atrophy patterns, including their spatial distribution, may explain remaining variability.
Approach:
- Utilized deep learning (Convolutional Neural Networks - CNNs) on whole-brain morphometry and lesion data from 231 chronic stroke patients.
- Compared CNN performance against Support Vector Machines (SVMs) for predicting severe aphasia.
- Investigated the ability of CNNs to identify individualized spatial patterns of atrophy.
Key Points:
- CNNs significantly outperformed SVMs in predicting aphasia severity.
- CNNs identified widely distributed atrophy patterns, unlike SVMs which focused near the lesion.
- Distinct, consistent atrophy patterns were identified, implicating specific brain networks.
Conclusions:
- Three-dimensional atrophy patterns are directly linked to aphasia severity.
- Deep learning offers potential for improved neuroimaging-based prognostication of behavioral outcomes.
- Analyzing spatial dependencies in neuroimaging data can enhance predictions of post-stroke recovery.
