Abstract
This article combined medical images and biomechanical data to construct a sports injury prediction model, solving the issues of incomplete data acquisition and analysis, lack of comprehensive prediction models, insufficient consideration of individual differences, lack of real-time monitoring and preventive measures, and limited technical means in traditional aerobics sports injury research. It studied the collection of a large number of MRI and CT images, using median filtering and Gaussian filtering for denoising processing, and image enhancement through histogram equalization. Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) can be used to extract and fuse features from preprocessed images and biomechanical data. The motion capture system and force platform collect joint angles, muscle strength, motion trajectory and other data, and integrate medical images and biomechanical data through multimodal fusion methods. The constructed prediction model is based on the extraction and integration of key features, combined with individual differences to provide personalized injury prevention recommendations. The system has developed a real-time monitoring function, which collects data in real time through sensors and wearable devices, conducts response time testing using the performance testing tool Apache JMeter, and evaluates the accuracy of warnings through a confusion matrix. The experimental results show that the constructed model achieves an accuracy of 85%, a precision of 82%, a recall rate of 90%, and an F1 value of 86%, all of which are superior to traditional methods in various indicators. The system designed in this article improves the accuracy and real-time performance of predicting injuries in aerobics, providing reliable prevention and monitoring methods for athletes.