Abstract:Objective Ultrasound imaging can be used to monitor bone healing after fracture. It has the advantages of simplicity, rapidity, non-radiation, and real-time dynamic observation of fracture healing. It is increasingly valued and developed in clinical applications. Deep learning is an essential branch of artificial intelligence widely used in the medical field. Methods After fracture surgery, 134 patients were selected for ultrasonic imaging at 2 weeks, 4 weeks, 8 weeks, and 12 weeks after surgery. Among them, 104 patients were selected for ultrasonic imaging to establish a prediction model for ultrasonic deep learning, and then 30 patients were used for imaging to verify and improve the model. Finally, 30 patients were selected for ultrasonic imaging and X-ray imaging at 2 weeks, 4 weeks, 8 weeks, and 12 weeks after fracture surgery for the same patient; test the ultrasonic imaging prediction model, compare the ultrasonic imaging and X-ray imaging manual reading, and evaluate the ultrasonic imaging prediction model by accuracy, loss curve, F1 score value, roc curve, AUC value, etc. Results The accuracy of the ultrasonic imaging deep learning prediction model in the test set was 0.73. In the test set, the predicted AUC values for 2 weeks, 4 weeks, 8 weeks, and 12 weeks after fracture surgery were 0.82, 0.90, 0.86, and 0.67, respectively. The accuracy of ultrasonic imaging manual film reading was 0.7, and the AUC values for the four periods were 0.76, 0.81,0.81, and 0.86, respectively. X-ray imaging manual film reading accuracy was 0.7, and the AUC values for the four periods were 0.86, 0.76, 0.81, and 0.84, respectively. Conclusion Based on the artificial intelligence framework, a deep learning prediction model is constructed by preprocessing the ultrasonic imaging after fracture surgery to predict fracture healing, and its application value is discussed. The prediction model of ultrasonic imaging based on deep learning can be used as an effective monitoring method for bone healing after fracture surgery.

Deep Learning Approach for Automatic Wrist Fracture Detection Using Ultrasound Bone Probability Maps

Critical evaluation of deep neural networks for wrist fracture detection

Domain Specific Transporter Framework to Detect Fractures in Ultrasound

Detecting pediatric wrist fractures using deep-learning-based object detection

Enhancing Wrist Fracture Detection with YOLO

Deep Learning Model for Automatic Identification and Classification of Distal Radius Fracture

Learning from the few: Fine-grained approach to pediatric wrist pathology recognition on a limited dataset

Deep learning performance compared to healthcare experts in detecting wrist fractures from radiographs: A systematic review and meta-analysis

Fracture Detection in Wrist X-ray Images Using Deep Learning-Based Object Detection Models

Artificial intelligence detection of distal radius fractures: a comparison between the convolutional neural network and professional assessments

Deep neural network improves fracture detection by clinicians

Assessment of a Deep Learning Algorithm for the Detection of Rib Fractures on Whole-Body Trauma Computed Tomography

[Study on automatic and rapid diagnosis of distal radius fracture by X-ray]

Automatic Hip Fracture Identification and Functional Subclassification with Deep Learning

Automated universal fractures detection in X-ray images based on deep learning approach

Fractured Elbow Classification Using Hand-Crafted and Deep Feature Fusion and Selection Based on Whale Optimization Approach

Improved Automated Quality Control of Skeletal Wrist Radiographs Using Deep Multitask Learning

Application of deep learning ultrasound imaging in monitoring bone healing after fracture surgery

Skeletal Fracture Detection with Deep Learning: A Comprehensive Review

Bone Fracture Detection Using Deep Supervised Learning from Radiological Images: A Paradigm Shift