Observer-Blinded Comparison of Virtual and Manual Finger ROM in AI-Based 3-Dimensional Hand Models.

PURPOSE: To examine the validity, agreement, and intra- and inter-rater reliability of an artificial intelligence (AI)-based 3-dimensional hand modeling system with virtual goniometry for assessing finger joint range of motion (ROM) in individuals with postinjury movement limitations. METHODS: This cross-sectional, observer-blinded validation study included individuals with persistent finger ROM limitations following hand injury. Finger joint ROM was assessed using both manual goniometry and AI-based 3-dimensional hand modeling with virtual goniometry. Measurements were obtained for the metacarpophalangeal (MCP), proximal interphalangeal (PIP), distal interphalangeal (DIP), and thumb interphalangeal (IP) joints. The agreement between measurement methods and the consistency of repeated measurements across and within raters were evaluated. RESULTS: Measurements obtained using AI-based 3-dimensional hand modeling demonstrated strong consistency with manual goniometry for MCP, PIP, and DIP joint flexion. Agreement among methods showed no systematic bias for these joints. Intra- and inter-rater reliability were high for finger joint flexion measurements. Greater variability was observed in thumb IP flexion and total finger extension measurements compared with other joint motions. CONCLUSIONS: AI-based 3-dimensional hand modeling with virtual goniometry provides valid and reliable measurements for assessing MCP, PIP, and DIP joint flexion following hand injury. The increased variability observed in thumb IP flexion and finger extension measurements indicates that these outcomes should be interpreted with caution when used in remote or technology-assisted clinical assessments. TYPE OF STUDY/LEVEL OF EVIDENCE: Diagnostic IIb.