Enhanced deep residual network for bone classification and abnormality detection.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Orthopedic disease diagnosis relies heavily on X-ray imaging.
• Accurate and rapid detection of bone abnormalities is crucial for effective patient treatment.
• Current diagnostic methods can be time-consuming and prone to errors.

Purpose of the Study:

• To develop a two-stage deep learning framework for bone classification and abnormality detection using X-rays.
• To enhance the speed and accuracy of orthopedic disease diagnosis.
• To reduce the incidence of false diagnoses in clinical practice.

Main Methods:

• A two-stage deep learning approach was implemented.
• Stage 1: Bone classification using ResNeXt50 to mitigate bone type variations.
• Stage 2: Seven anomaly detectors trained per bone type for abnormality detection, employing techniques like data augmentation and focal loss to optimize performance and prevent overfitting.

Main Results:

• The bone classification stage achieved 96.69% accuracy, 96.69% sensitivity, 99.46% specificity, and 96.42% F1 score.
• The abnormality detection stage yielded 84.15% accuracy, 84.15% sensitivity, 87.50% specificity, 84.10% F1 score, 0.72 Cohen's Kappa, and 0.90 AUC.
• Experiments were conducted on the MURA dataset, the largest available for bone abnormality detection.

Conclusions:

• The proposed two-stage deep learning framework demonstrates superior effectiveness compared to other convolutional neural network models.
• The framework achieved improved metrics for bone classification and abnormality detection, including accuracy, sensitivity, specificity, F1 score, Cohen's Kappa, and AUC.
• This approach offers a promising solution for accelerating orthopedic diagnosis and improving diagnostic accuracy.