Correlation between High-frequency Ultrasound Imaging Features Using Improved Convolutional Neural Network Model and the Efficacy of Modified Tongluo Qushi Chubi Decoction in the Treatment of Rheumatoid Arthritis

An improved convolutional neural network (CNN) model was constructed in this study for the automated analysis of high-frequency ultrasound (HFUS) images of patients with rheumatoid arthritis (RA) and to explore the correlation between its imaging features and the therapeutic efficacy of modified Tongluo Qushi Chubi Decoction. A prospective, randomized, double-blind, placebo-controlled trial was conducted, with 106 patients with active RA being divided into the traditional Chinese medicine (TCM) group (conventional Western medicine treatment + modified Tongluo Qushi Chubi Decoction) and the control group (conventional Western medicine treatment + placebo), each consisting of 53 cases. The improved CNN model was used to automatically segment and grade the ultrasound images, analyze the changes in synovial hypertrophy (gray-scale ultrasound, GSUS) and power Doppler ultrasound (PDUS) signals, and combine clinical indicators for evaluation. The results showed that the automated scoring of the improved CNN model was highly consistent with manual assessment (weighted Kappa > 0.80). After treatment, the TCM group showed significantly greater improvements than the control group in disease activity score (DAS28) reduction, American College of Rheumatology (ACR) response rate, health assessment questionnaire (HAQ) score improvement, and inflammatory indicators (all P < 0.05). Radiologically, the improvements in GSUS and PDUS scores were also more significant in the TCM group (both P < 0.001). ΔPDUS and ΔGSUS were significantly positively correlated with clinical indicators such as ΔDAS28 and Δ erythrocyte sedimentation rate (ESR) (r=0.58–0.75), with stronger correlations in the TCM group. Multivariate regression analysis indicated that ΔPDUS, ΔGSUS, and TCM treatment were all independent predictors of ΔDAS28 (all P < 0.01). In summary, the improved CNN model can achieve objective and automated evaluation of RA ultrasound imaging; modified Tongluo Qushi Chubi Decoction treatment can significantly improve patients’ clinical, inflammatory, and radiological manifestations, with radiological changes closely related to clinical efficacy, providing objective radiological evidence for the treatment of RA with TCM.