AI landmark versus conventional ultrasound-guided injections for adhesive capsulitis: a prospective, propensity score-matched, superiority study

Abstract

Aim: To investigate the artificial Intelligence (AI) landmark for guiding rotator cuff interval injections for adhesive capsulitis (AC).

Material and methods: One hundred fifty-five patients undergoing AI landmark-guided injections were prospectively enrolled (AI cohort). A 1:1 propensity-score matching was conducted for ultrasound (US) cohort receiving US-guided injections (n=149). The primary endpoint was shoulder pain and disability index (SPADI) scores at 1-month post-injections, with a defined superiority margin of a 10-point decrease. Secondary outcomes included needle insertions, procedure time, range off motion, quality of life (QoL) and adverse events.

Results: SPADI scores at 1-month follow-up was 32.81±18.16 and 44.53±11.74 in two cohorts with a mean difference of -11.72 (95%CI: -17.95, -10.48), confirming the superiority (p=0.025). Higher means were observed at 1-week and 3-month post-injections (p=0.018 and 0.033). Better improvements were observed in forward flexion (158.77±13.12 vs. 145.97±11.49, p=0.016), abduction (78.98±15.96 vs. 61.09±17.09, p=0.007) and external rotation (63.13±19.97 vs. 50.19±13.85, p=0.019) in AI cohort compared to US cohort at 1-week post-injections. AI cohort had fewer needle insertions, shorter procedure times and better QoL scores. Minor adverse events were observed.

Conclusions: AI approach facilitated the identification of shoulder landmarks and demonstrated superiority in relieving shoulder pain and improving function compared to ultrasound method.