Comprehensive deep learning-assisted multi-condition analysis of knee MRI studies improves resident radiologist...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Increasing MRI utilization places pressure on radiologists.
• There is a need for comprehensive AI models for efficient knee MRI analysis.
• Automated analysis of multi-tissue, multi-condition knee MRI is crucial.

Purpose of the Study:

• To develop a deep-learning model for automated knee MRI analysis.
• To assess the clinical potential and diagnostic performance of the AI model.
• To evaluate the model's impact on radiology resident performance.

Main Methods:

• A 3D slice transformer network was trained on 3121 knee MRI studies.
• The model was evaluated using cross-validation and an external test set (448 studies).
• Diagnostic performance was compared between residents with and without AI assistance.

Main Results:

• The AI model achieved an AUC of at least 0.85 for 8 conditions and 0.75 for 18 conditions.
• Model assistance improved accuracy and sensitivity for inexperienced residents.
• AI assistance increased sensitivity and reduced reading times by 10% for experienced residents.

Conclusions:

• The deep-learning model demonstrates robust performance across diverse knee conditions.
• AI assistance enhances diagnostic performance and efficiency of radiology residents.
• Future work should focus on subtle or rare conditions for comprehensive joint assessment.