Deep learning in bone imaging diagnosis

Abstract

BACKGROUND: Deep learning methods have made breakthrough progress in the field of bone imaging diagnosis. They have overcome the problems of easy misdiagnosis and low efficiency in traditional bone imaging diagnosis methods, and are conducive to the popularization of intelligent diagnosis methods in orthopedics. OBJECTIVE: To review the application, advantages and disadvantages of deep learning in the diagnosis of common bone diseases. METHODS: Literature published from January 2021 to June 2025 on deep learning-assisted skeletal image diagnosis was retrieved from CNKI, WanFang, PubMed, and Web of Science databases. Chinese and English search terms included “artificial intelligence, deep learning, machine learning, computer-aided diagnosis, skeletal imaging, fracture, bone tumor, osteoporosis, osteoarthritis, synovitis, spinal, cartilage, classification, detection, segmentation.” According to the inclusion criteria, 76 articles were finally included in this review. RESULTS AND CONCLUSION: Deep learning models have become powerful tools for bone imaging diagnosis and are gradually gaining recognition from clinicians, improving the efficiency of bone imaging diagnosis. Deep learning technology uses its image feature capture ability to help improve the clinical diagnosis of fractures, bone tumors, osteoporosis, osteoarthritis, synovitis, and spinal lesions, providing a reference for clinical decision-making. Although deep learning diagnostic applications have great potential, they are prone to insufficient model generalization and rely heavily on large amounts of annotated data, which reduces model credibility and hinders clinical translation. Future research should focus on improving the robustness and generalization of deep learning models. In summary, deep learning has certain reference value in clinical bone imaging diagnosis.

1. Introduction

Bone diseases are major public health problems affecting human health worldwide. According to statistics, about 520 million people suffer from osteoarthritis globally each year, osteoporosis causes about 9 million fragility fractures, and bone tumors have a high mortality rate. The structure and function of bones are the basis for maintaining human activities and are closely related to the body's mobility. Common clinical conditions, including fractures, bone tumors, osteoporosis, osteoarthritis, synovitis, and spinal lesions, seriously affect patients' normal life and impose a heavy burden on the healthcare system.

The diagnosis of bone diseases is currently based on medical imaging techniques such as X-ray, CT, and MRI, but the large amount of image data and the complexity of lesions can easily limit the diagnostic efficiency and accuracy of physicians. Therefore, it is of great significance to study efficient intelligent diagnostic methods for bone imaging.

With the rapid development of medical artificial intelligence, deep learning has been able to capture key features in complex bone images, thereby enabling complex lesion analysis and providing new ideas for the field of medical image analysis. More and more studies have proven that deep learning models based on convolutional neural networks are developing rapidly in the field of bone disease diagnosis, significantly improving the diagnostic efficiency and accuracy. However, systematic reviews on the application of deep learning in bone imaging diagnosis are relatively insufficient, lacking an overall analysis of multiple bone disease diagnostic applications. This article addresses the diagnostic problems of traditional bone imaging, summarizes the research progress of advanced deep learning applications, and provides a reference for clinical personnel.