Multimodal Deep Learning for Grading Carpal Tunnel Syndrome: A Multicenter Study in China
Xiaochen Shi1, Tianxiang Yu2, Yu Yuan3
1Department of Trauma and Orthopedics, Peking University People's Hospital, No.11, Xizhimen South Street, Xicheng District, Beijing 100044, PR China (X.S.).
Academic Radiology
|March 29, 2025
Summary
A new deep learning model, CTSGrader, effectively grades carpal tunnel syndrome severity by combining ultrasound images and clinical data. This AI tool shows potential to aid doctors in clinical decision-making.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Deep learning (DL) models for carpal tunnel syndrome (CTS) severity grading using ultrasound (US) are limited.
- Advancing CTS diagnosis requires integrating multimodal data for improved accuracy.
Purpose of the Study:
- To develop and validate a joint deep learning model (CTSGrader) integrating clinical information and multimodal US features for CTS severity grading.
- To compare the performance of the joint DL model against a US-only DL model and human radiologists.
Main Methods:
- A retrospective dataset was used for training and internal validation, with external validation from prospective and cross-vendor datasets.
- A joint DL model (CTSGrader) was created using sonographic features (CSA, echogenicity, nerve appearance, vascularity) and clinical data.
- Model performance was evaluated against electrophysiological results and compared to radiologists, with and without AI assistance.
Main Results:
- CTSGrader demonstrated high diagnostic performance with AUCs ranging from 0.897 to 0.951 across validation sets.
- The joint DL model significantly outperformed the US-based DL model and junior radiologists, performing comparably to senior radiologists.
- AI assistance improved the diagnostic accuracy of radiologists.
Conclusions:
- The developed joint DL model (CTSGrader) shows superior performance compared to single-modality models for CTS grading.
- The AI-aided approach holds significant potential to support clinical decision-making in grading CTS severity.
