J Med Internet Res. 2025 Sep 3;27:e63308. doi: 10.2196/63308.
ABSTRACT
BACKGROUND: Shoulder pain is a highly prevalent musculoskeletal disorder that severely compromises patients' quality of life. The Constant-Murley Scale (CMS) is a well-established method for shoulder function evaluation. However, the necessity of clinician involvement constrains its utility in continuous monitoring. Recent improvements in human pose estimation and inertial sensors provide possibilities for automated functional assessment.
OBJECTIVE: This study introduces an automated CMS assessment system that can provide objective measuring results using movement images and inertial sensor data (Mobile Constant) and aims to evaluate its reliability by comparison with standard results from human raters.
METHODS: The Mobile Constant system integrated subjective symptom questionnaires, range-of-motion analysis, and strength assessment. Patients presenting with shoulder concerns were enrolled consecutively, with movement images and inertial sensor data collected from each participant. The dataset was structured as follows: patients recruited from February to November 2022 at our hospital formed the training set, those enrolled between December 2022 and February 2023 served as the internal validation set, and patients recruited from April to July 2025 at an independent hospital constituted the external validation set. Gold standard assessments were determined independently by 2 raters using standardized protocols. Six machine learning models (logistic regression, k-nearest neighbors, decision tree, support vector machine, random forest, and adaptive boosting) were developed. The reliability of the system was determined by comparison with human raters using differences, Cohen κ coefficients, and intraclass correlation coefficients (ICCs). Agreement across human raters was also evaluated by comparison between 4 independent clinicians.
RESULTS: Data from 141 patients with shoulder pain and stiffness were collected (training set: n=83, 58.9%; internal validation set: n=28; 19.9%; external validation set: n=30, 21.3%). For range-of-motion analysis, the Mobile Constant system showed fair to substantial reliability, achieving κ coefficients ranging from 0.498 to 0.819 and ICCs ranging from 0.898 to 0.956 in the internal validation set. In the external validation set, κ coefficients ranged from 0.198 to 0.699, and ICCs ranged from 0.584 to 0.922. For abduction strength assessment, the k-nearest neighbors model demonstrated substantial reliability, yielding a κ coefficient of 0.707 and an ICC of 0.759 in internal validation and higher agreement in external validation (κ=0.809; ICC=0.906).
CONCLUSIONS: The self-reported method for shoulder function evaluation demonstrated substantial agreement with experienced human raters. The proposed system enabled reliable patient-conducted assessment using mobile phone-integrated cameras and inertial sensors and exhibited strong potential for remote monitoring.
PMID:40902145 | DOI:10.2196/63308