ATHLETE MUSCULOSKELETAL INJURY RECOGNITION BASED ON TRANSFER LEARNING MRI SCANNING

Abstract

Currently, the prevalence of musculoskeletal injuries in high-level competitive athletes is high, and the injuries mainly appear in the neck and lower back. Long training years, looking down during training, and engaging in real-time combat sports are the main risk factors for the occurrence of neck disorders, and therefore, a scientific method is needed to identify signs of musculoskeletal injuries in athletes' MRI scans for early prevention and early intervention. Therefore, in this paper, a network model based on the lightweight network MobilenetV2 is proposed based on the transfer learning method, which ensures high classification accuracy while completing the recognition task with less number of parameters and computation. To address the problem that deep learning models require a large amount of data for training and patient images are difficult to obtain, this paper proposes the RS algorithm, which masks the local features of the 2DMRI images, augments the dataset, and combines it with the two proposed deep migratory learning classification models, TLV2 and TLV2C, in which TLV2 is the MobileNetV2 that fixes the shallow layer and initializes the deep layer after training on the Image Net dataset. network, initialize the deep network and re-train it under MRI dataset. TLV2C redesigns the classifier on the basis of TLV2. The experimental results show that the algorithm proposed in this paper solves the problem of insufficient MRI data, and at the same time is able to identify the signs of musculoskeletal injuries in MRI scans of athletes