Deep Learning Model for Automatic Identification and Classification of Distal Radius Fracture

Area of Science:

• Radiology
• Artificial Intelligence
• Orthopedics

Background:

• Distal radius fractures (DRF) are common wrist injuries.
• Accurate and timely diagnosis is crucial for effective treatment.
• Current diagnostic methods can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate deep learning models for automatic segmentation, identification, and classification of DRF from wrist radiographs.
• To assess the performance of various deep learning architectures in analyzing wrist X-rays.

Main Methods:

• Utilized a dataset of 2240 anteroposterior wrist radiographs.
• Employed Unet and Fast-RCNN for automatic radiograph segmentation.
• Applied DenseNet121 and ResNet50 for DRF identification.
• Used DenseNet121, ResNet50, VGG-19, and InceptionV3 for DRF classification (Types A, B, C).
• Evaluated models using Area Under the Curve (AUC), accuracy, precision, and F1-score.

Main Results:

• Deep learning models demonstrated effective segmentation of wrist radiographs.
• DenseNet121 achieved an AUC of 0.941 for DRF identification.
• DenseNet121 achieved an AUC of 0.96 for classifying DRF types A, B, and C.
• 1440 out of 2240 participants had DRF, with varying types.

Conclusions:

• Deep learning models show significant potential for automated analysis of wrist radiographs.
• The DenseNet121 model is a promising tool for assisting clinicians in DRF diagnosis and classification.
• Automated analysis can improve efficiency and accuracy in interpreting wrist fractures.