AI assisted diagnosis using DEANet to improve correct diagnosis of iliac wing fracture and ischial spine fracture

Abstract

Background

Hip fractures—particularly those involving the iliac wing and ischial spine—are complex injuries that often require CT or MRI scans for accurate diagnosis. However, these imaging modalities are costly, time-consuming, and involve exposure to radiation or contrast-related risks. This study aims to develop an AI-based classification model, the Dynamic Efficient Attention Network, capable of automatically distinguishing between iliac wing and ischial spine fractures using pelvic X-ray images. The objective is to facilitate early diagnosis and reduce reliance on advanced imaging modalities.

Methods

The proposed method employs a dual-branch architecture that integrates EfficientNet-B0 with an Enhanced Depth-wise Separable Attention Block to enhance edge-region feature representation. The model was trained using pelvic X-ray images collected from China Medical University Hospital and evaluated based on accuracy, precision, recall, F1 score, and Intersection over Union.

Findings

The model achieved an accuracy of 85 % on the test dataset and demonstrated robust performance across all evaluation metrics. These findings suggest that the proposed method has the potential to function as a reliable AI-assisted diagnostic tool for the early and accurate classification of hip fractures, thereby supporting clinical decision-making and improving treatment planning.

Interpretation

Compared to existing approaches that rely on CT or MRI imaging, the proposed method demonstrates that advanced processing of X-ray images can yield clinically meaningful classification results. This underscores the potential of the proposed method as a cost-effective, efficient, and accessible diagnostic tool, especially in settings where access to advanced imaging modalities is limited.