Formal methods for early prediction of frozen shoulder severity and recovery: A multicentre prospective study

Background: Frozen Shoulder (FS) is a pathological state resulting in severe pain and global functional limitations. Although diagnosis is primarily clinical, traditional imaging often lacks prognostic utility. However, to date, only conventional human evaluation of imaging has been explored, while artificial intelligence remains uninvestigated. We aimed to synthesize magnetic resonance imaging (MRI)-derived quantitative radiomic features with clinical markers to predict baseline severity and three-month functional trajectories. Methods: This multicenter, prospective, single-cohort study included subjects diagnosed with FS who were followed for three months and recruited from four orthopedic and rehabilitation centers in Italy. The recruited subjects were adults of both sexes diagnosed with FS according to the international guidelines. Outcomes were quantified using the Shoulder Pain and Disability Index (SPADI) and the Disabilities of the Arm, Shoulder, and Hand (DASH) score. MRI data were also acquired for each subject. Radiomic analysis and Formal Methods were applied and correlated with clinical outcomes. Results: The cohort was characterized by a female predominance (55.6%), a core age demographic of 40–50 years (51.9%), and a high prevalence of sedentary professionals (70.4%). Data analysis revealed that specific textural classes, notably ‘Gray-Level Run Length Matrix’ and ‘Neighboring Gray Tone Difference Matrix’, served as potent biomarkers. These features effectively stratified patients by baseline severity and accurately forecasted clinical outcomes at the three-month follow-up. Conclusion: The synergy between Radiomics and Formal Methods demonstrates significant predictive power for both SPADI and DASH metrics. These results facilitate a transition toward more objective, generalizable models for the precision management of FS.