Enhancing X-ray-Based Wrist Fracture Diagnosis Using HyperColumn-Convolutional Block Attention Module.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Fractures impact a significant portion of the population, with radiography as the standard diagnostic method.
• Current deep learning models for fracture detection face challenges with annotation reliance and diagnostic precision.

Purpose of the Study:

• To enhance fracture detection accuracy in X-ray radiographs using machine learning.
• To integrate HyperColumn techniques and the Convolutional Block Attention Module (CBAM) with deep learning models.

Main Methods:

• Employed EfficientNet-B0 and DenseNet169 models integrated with HyperColumn and CBAM.
• Utilized Grad-CAM technology to refine heatmap focus and align with expert annotations.

Main Results:

• Both DenseNet169 and EfficientNet-B0 models showed accuracy improvements (approx. 0.69% and 0.70%).
• The HyperColumn-CBAM-DenseNet169 model achieved a notable AUC score increase from 0.8778 to 0.9145.
• Grad-CAM improved heatmap accuracy, reducing reliance on bounding box annotations.

Conclusions:

• The integration of HyperColumn and CBAM significantly enhances fracture detection in X-ray images.
• This approach offers potential for streamlined training and improved diagnostic accuracy in fracture detection.
• The method shows promise in overcoming limitations of traditional deep learning annotation requirements.