Designing an Intelligent Lesion Detection System Using Deep Architecture Neural Networks in the Lower Limb X-Ray Images

Abstract

Purpose: Diagnosis of musculoskeletal abnormalities is essential due to more than 1.7 billion people worldwide being affected by musculoskeletal disorders. In this study, we focus on diagnosing musculoskeletal abnormalities in the lower extremities within X-ray images by deep architecture neural networks.

Methods: Our dataset contains 61,098 musculoskeletal radiographic images, which includes 42658 normal images and 18440 abnormal images. Each image belongs to a single type of lower extremity radiography, including the toe, foot, ankle, leg, knee, femur, and hip joint. We proposed a new deep neural network architecture with two different scenarios that perform lower extremity lesion diagnosis functions with high accuracy. The core of the proposed method is a deep learning framework based on the Mask R-CNN and CNN. The model with the best results utilized the Mask R-CNN algorithm to generate the bounding box, followed by the CNN algorithm to detect the class based on that.

Results: The proposed model can detect different types of lower limb lesions by an AUC-ROC of 0.925, with an operating point of 0.859 sensitivity and a specificity of 0.893.

Conclusions: By comparing the different results, it can be concluded that the consecutive implementation of Mask R-CNN and CNN function better than Mask R-CNN and CNN separately.