This randomized controlled trial (RCT) evaluates the efficacy of a novel Artificial Intelligence (AI)-based digital therapeutic system compared to traditional outpatient Schroth exercises for the treatment of Adolescent Idiopathic Scoliosis (AIS). The intervention utilizes a smartphone application with computer vision technology to perform remote, personalized posture analysis and generate adaptive exercise prescriptions. The study aims to determine if this "Human-in-the-Loop" AI model demonstrates superior clinical outcomes in controlling spinal curve progression and improving treatment adherence compared to standard conservative care over a 6-month period.

Background: Adolescent Idiopathic Scoliosis (AIS) requires long-term conservative management to prevent curve progression. While Physiotherapeutic Scoliosis Specific Exercises (PSSE), such as the Schroth method, are the gold standard, their efficacy is often limited by accessibility barriers and suboptimal adherence.

Study Design: This is a parallel-group, single-blind randomized controlled trial. A total of 300 eligible adolescents (aged 10-18, Cobb angle 10-30°, Risser sign 0-2) will be randomized in a 1:1 ratio into an Intervention Group or a Control Group.

Interventions:

Intervention Group (AI-DTx): Participants utilize a smartphone application to capture standardized photos every two weeks. An AI algorithm (based on MediaPipe) extracts skeletal landmarks to classify curve patterns and assign personalized exercise modules. The system features a "Check-and-Adjust" matrix to adapt exercise intensity based on metric changes. A Clinical Research Assistant (Human-in-the-Loop) verifies all prescriptions for safety before release.

Control Group (Standard Care): Participants receive standard outpatient Schroth therapy, involving daily home exercises and monthly supervised clinic visits, following SOSORT guidelines.

Objectives: The primary objective is to evaluate the absolute change in the major curve Cobb angle from baseline to six months. Secondary outcomes include the Angle of Trunk Rotation (ATR), trunk appearance perception (TAPS), quality of life (SRS-22r), and adherence rates. The study hypothesizes that the AI-driven intervention will demonstrate superior efficacy and adherence compared to traditional outpatient care.

Inclusion Criteria:

1. Diagnosis of Adolescent Idiopathic Scoliosis (AIS).
2. Age between 10 and 18 years (inclusive).
3. Cobb angle of the major curve between 10° and 30°.
4. Risser sign 0-2 (indicating skeletal immaturity).
5. Ability to operate a smartphone and follow instructions for the digital therapeutic application.
6. Patient and legal guardian provide written informed consent.

Exclusion Criteria:

1. Non-idiopathic scoliosis (e.g., congenital, neuromuscular, or syndromic etiology).
2. History of prior spinal surgery or currently scheduled for surgery.
3. Contraindications to exercise therapy (e.g., cardiac or respiratory insufficiency).
4. Severe cognitive or psychiatric disorders that prevent compliance with the protocol.
5. Currently receiving other scoliosis-specific treatments (e.g., bracing) outside of the study protocol (unless stratified).

De-identified individual participant data underlying the published results, including baseline demographics, Risser sign, Cobb angle, Angle of Trunk Rotation, TAPS scores, SRS-22r scores, and adherence metrics for all randomised participants, will be made available. All data will be anonymised prior to sharing; no facial images or directly identifying information will be released.

Beginning 9 months after publication of the primary results article and ending 36 months thereafter.

Access will be granted to qualified researchers whose proposed use has been approved by an independent review committee for the purpose of individual participant data meta-analyses or methodologically sound secondary analyses. Requests should be directed to the corresponding author (penny@link.cuhk.edu.hk) and must include a detailed research proposal, a statistical analysis plan, and evidence of ethical approval from the requester's institution. Data will be shared through a secure data transfer platform following the execution of a signed Data Use Agreement.

Due to the visible nature of the exercise intervention, it is not possible to blind participants or care providers. To minimize detection bias, the primary outcome (Cobb angle) will be assessed by independent radiologists who are strictly blinded to group allocation and the timing of the radiographs.