Bone Fracture Classification using Transfer Learning

TL;DR

The paper tackles automatic bone fracture classification from X-ray images to reduce missed fractures and delays. It employs transfer learning with EfficientNet-B6 on the FracAtlas dataset, supported by a concise training loop and a targeted preprocessing pipeline. The results show near 100% training accuracy and about 96.83% test accuracy within seven epochs, with high precision, recall, and AUC ROC, indicating robust clinical potential. The work highlights the importance of data quality and responsible training, and suggests extending the approach to precise fracture localization using bounding boxes and addressing class imbalance in future work.

Abstract

The manual examination of X-ray images for fractures is a time-consuming process that is prone to human error. In this work, we introduce a robust yet simple training loop for the classification of fractures, which significantly outperforms existing methods. Our method achieves superior performance in less than ten epochs and utilizes the latest dataset to deliver the best-performing model for this task. We emphasize the importance of training deep learning models responsibly and efficiently, as well as the critical role of selecting high-quality datasets.

Figures (3)

• Figure 1: We identified using transformations did not help us identify fractures neither helped us to get good results. Thus, making our training loop simpler and easier to learn/generalize on raw xray-images
• Figure 2: Metrics
• Figure 3: Results From Deployed Model