Interpretable Multi-Label Classification for Tibiofibula Fracture 2D CT Images with Selective Attention and Data...

Area of Science:

• Orthopedics
• Radiology
• Artificial Intelligence in Medicine

Background:

• Tibiofibula fractures are common across all age groups, often leading to frequent postoperative complications.
• Current classification methods for tibiofibula fractures can be challenging due to varied locations and uneven fracture type distribution.
• Accurate and rapid fracture classification is crucial for effective clinical management.

Purpose of the Study:

• To develop an interpretable deep learning model for multi-label classification of tibiofibula fractures using 2D CT images.
• To address challenges of limited sample size and class imbalance in fracture datasets.
• To provide visual interpretation of the model's classification decisions.

Main Methods:

• A deep learning model was developed for multi-label classification of tibiofibula fractures from 2494 2D CT images.
• The model utilized data augmentation techniques to handle limited data and class imbalance.
• Saliency maps generated by Grad-CAM++ provided visual interpretation for each classified fracture type.

Main Results:

• The proposed deep learning model achieved a mean average precision (mAP) of 95.71% for tibiofibula fracture classification.
• The model demonstrated effectiveness in classifying fractures despite challenges like limited sample size and uneven fracture distribution.
• Visual interpretation through saliency maps confirmed the model's reliable decision-making process.

Conclusions:

• The developed deep learning model offers an accurate and interpretable method for classifying tibiofibula fractures from CT scans.
• Saliency map-based visual interpretation enhances trust and allows for verification of the model's diagnostic reasoning.
• This approach has the potential to significantly assist physicians in diagnosing and managing tibiofibula fractures.