KneeXNet-2.5D: a clinically-oriented and explainable deep learning framework for MRI-based knee cartilage and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Accurate knee cartilage and meniscus segmentation in MRI is crucial for diagnosing conditions like osteoarthritis.
• Manual segmentation is labor-intensive, subjective, and not clinically feasible.
• Existing methods lack efficiency and interpretability for widespread clinical adoption.

Purpose of the Study:

• To introduce KneeXNet-2.5D, a deep learning framework for efficient and explainable knee MRI segmentation.
• To improve the accuracy and robustness of AI models for cartilage and meniscus segmentation.
• To facilitate clinical integration and open scientific research through public data and code release.

Main Methods:

• Developed a 2.5D deep learning architecture to capture inter-slice context in sagittal knee MRIs.
• Implemented targeted data augmentation, including synthetic noise injection, to enhance model robustness.
• Integrated an entropy-based AI explainability method for model transparency.

Main Results:

• Achieved high segmentation accuracy for knee cartilage and meniscus using the 2.5D approach.
• Demonstrated computational efficiency and reduced resource consumption suitable for low-resource environments.
• Validated clinical relevance and anatomical fidelity through expert orthopedic surgeon review.

Conclusions:

• KneeXNet-2.5D offers a clinically viable solution for automated knee MRI segmentation.
• The framework's explainability and efficiency support its integration into musculoskeletal imaging workflows.
• Publicly releasing the dataset, code, and models promotes further research and development in AI for medical imaging.