AI-Assisted Ultrasound Therapy for Adhesive Capsulitis

Overview

This study aims to integrate artificial intelligence with ultrasound imaging to investigate the dynamic biomechanical characteristics of adhesive capsulitis and to compare the therapeutic effectiveness of three ultrasound-guided hydrodilatation techniques: rotator interval injection, dual-target injection, and posterior glenohumeral injection. In the observational phase, healthy participants and patients with adhesive capsulitis will undergo static and dynamic ultrasound evaluations to quantify subacromial motion and the minimal vertical acromiohumeral distance (mvAHD) during shoulder abduction. A Faster R-CNN model will automatically identify the acromion and greater tuberosity to extract motion trajectories and frequency-based features. In the randomized clinical trial phase, patients will be assigned to one of the three hydrodilatation techniques to compare improvements in pain, function, and range of motion at 6 and 12 weeks. All participants will subsequently enter a one-year follow-up to document recurrence, defined as the need for repeat intervention. Clinical characteristics, static sonographic parameters, dynamic motion metrics, and short-term treatment responses will be incorporated into machine-learning models to predict long-term outcomes, including recurrence risk. Eligible patients who decline randomization may consent to join a parallel observational cohort. They will receive standard-of-care treatment (conservative or interventional) based on their preference, and undergo the exact same baseline, Week 6, and Week 12 clinical and ultrasound assessments as the randomized trial participants. The ultimate goal is to establish an AI-assisted predictive framework that enables individualized risk stratification, early identification of poor responders, and optimization of injection strategies. This model is expected to improve clinical decision-making, enhance treatment precision, and contribute to higher-quality patient care.