Clinically Interpretable Bone Fracture Detection Using EfficientNet-B0 and XAI: A Deep Learning Approach for Radiographic Image Analysis

Abstract

The severity and complexity of bone fracture should be diagnosed quickly and correctly to facilitate the process of making clinical decisions and treating a patient in a timely manner. This is essential to be diagnosed early and treated early. This possibility has emerged due to advances that have been made in the field of deep learning in the recent past which hint at the possibility of automating this detection process. This study introduces a DL model which relies on EfficientNet-B0 and entirely incorporated with Explainable AI (XAI) methods, on bone X-ray images as an example of fracture identification. The model that was developed reported a high level of accuracy of 99.93% which outstripped numerous currently developed models. The results were compared with standard metrics and Grad-CAM visualization of the results indicates important regions of bones that determine the model predictions. It is a sound and simple model that is able to encourage usage in clinical settings and result in the accurate diagnosis and further informed treatment decisions due to the shortened delays. Future work will focus on validating across multiple datasets, extending to multiclass classification including fracture type and location, and involving expert validation of explainability results.

Ähnliche Arbeiten