Deep Learning Enhances Weightbearing CT Detection of Lisfranc Instability: A FIXUS-AI Ankle Insight 3D Algorithm.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning (DL) shows promise in musculoskeletal pathology detection.
• Weightbearing CT (WBCT) enhances diagnostic accuracy for subtle Lisfranc instability.
• The study explores DL's impact on WBCT for Lisfranc instability diagnosis.

Purpose of the Study:

• To investigate the efficacy of DL algorithms applied to WBCT images for diagnosing isolated Lisfranc instability.
• To compare the performance of different DL models in detecting Lisfranc injuries.

Main Methods:

• Evaluated 280 WBCT scans (140 cases, 140 controls) with an 80:10:10 train-validation-test split.
• Developed and trained three DL models: 3D-CNN, CNN-LSTM, and differential CNN-LSTM.
• Assessed model performance using sensitivity, specificity, accuracy, F1-score, and ROC curve analysis.

Main Results:

• Differential CNN-LSTM (Model 3) achieved an F1-score of 0.99, significantly outperforming the 3D-CNN (0.72) and CNN-LSTM (0.92).
• No significant baseline differences were observed between the case and control groups.
• The developed DL models demonstrated high diagnostic performance for Lisfranc instability.

Conclusions:

• A DL model was successfully developed for 3D WBCT-based Lisfranc injury detection with excellent accuracy.
• DL integration shows potential to improve diagnostic accuracy for Lisfranc instability.
• Further research with larger datasets and external validation is recommended.