Automated Detection Model Based on Deep Learning for Knee Joint Motion Injury due to Martial Arts

Abstract

Objective. Develop a set of knee joint martial arts injury monitoring models based on deep learning, train and evaluate the model's effectiveness. Methods. This paper mainly collects knee MRI images of 1546 patients with knee joint martial arts injuries from 2015 to 2020. Through manual annotation, the data set is divided into six categories: meniscus injury, tendon injury, ligament injury, epiphyseal cartilage injury and synovial joint capsule loss. The human knee collaborative MRI image database is established, and the data set is divided into the training and validation sets. And test set. Establish a deep neural network, train the model using the training set and validation set, locate the knee joint injury location, and classify the specific injury type. The model's validity was validated using the test set, and the model's sensitivity, specificity, and mean accuracy for detecting lesions were evaluated. Results. In the test set, the accuracy of meniscus injury, tendon injury, ligament injury, bone and bone cartilage injury and synovial joint capsule injury were 83.2%, 89.0%, 88.0%, 85.9%, 85.6% and 83.5%, respectively, and the overall average accuracy value was 86.0%. The sensitivity and specificity of the model were 91.3% and 87.3%, respectively. Conclusion. The application of the deep learning method in the classification and detection of knee joint martial arts injuries can significantly improve the diagnosis effect, reduce the diagnosis time and misdiagnosis rate, and provide decision support for surgery.