REHABILITATION EFFICACY OF DEEP LEARNING-BASED SPORTS TRAINING IN PATIENTS WITH CHRONIC LUMBAR DISC HERNIATION

Chronic lumbar disc herniation is a highly prevalent degenerative spinal disease. Traditional rehabilitation training has limitations such as subjective evaluation and homogeneous protocols, which affect the rehabilitation outcomes. Deep learning technology offers new possibilities for precise rehabilitation, but its clinical value needs to be verified through rigorous design. Objective: To construct a sports training system based on deep learning and explore its rehabilitation effects on patients with chronic lumbar disc herniation through a randomized controlled trial, achieving quantitative assessment and personalized intervention. An improved stacked hourglass network combined with a hybrid attention mechanism was used to build the pose estimation algorithm, and a spatiotemporal feature decoupling network was employed to recognize 12 types of rehabilitation movements. A total of 352 patients were randomly divided into the experimental group ([Formula: see text], using the intelligent system) and the control group ([Formula: see text], using conventional rehabilitation training). Both groups received interventions for 8 weeks and were followed up for 12 months. The effects were analyzed using independent samples t-test and paired samples t-test. In terms of technical performance, the localization accuracy of the lumbar key segments reached 92.6% ± 2.3%, the overall accuracy of motion recognition was 96.3% ± 1.5%, and the inference speed was 32.6 ± 1.8 frames per second, all of which were significantly better than traditional methods and open-source tools such as OpenPose (all [Formula: see text]). In terms of clinical outcomes, after 8 weeks of intervention, the VAS score in the experimental group decreased from 6.9 ± 1.4 to 2.3 ± 0.9 (improvement of 4.6 ± 1.2), and the ODI index decreased from 42.8 ± 8.1 to 18.7 ± 6.2 (improvement of 24.1 ± 7.5), both of which were significantly better than those in the control group (all [Formula: see text]). The 12-month follow-up showed that the recurrence rate in the experimental group (5.1%) was lower than that in the control group (15.3%, [Formula: see text]) and the training compliance was 89.7% ± 8.2%. The sports training system based on deep learning can accurately quantify rehabilitation movements, significantly improve patients’ pain and functional status, and achieve stable long-term effects, providing an effective solution for the intelligent rehabilitation of chronic lumbar disc herniation.