Deep learning assessment of visualization in rotator cuff repair: a secondary analysis of a randomized controlled trial comparing epinephrine irrigation and intravenous tranexamic acid.

Abstract

Background: This study aimed to compare intraoperative visualization between diluted epinephrine (EPN) irrigation and preoperative intravenous tranexamic acid (TXA) during the glenohumeral and subacromial phases of rotator cuff repair.

Methods: Eighty-two patients with rotator cuff tears were prospectively randomized (1:1) to receive either 0.33 mg/L EPN irrigation without TXA (EPN group) or preoperative intravenous TXA (15 mg/kg) without EPN irrigation (TXA group). Visual clarity was evaluated using a pretrained deep learning model with a 3-point grading scale across three stages: intra-articular assessment, subacromial debridement, and rotator cuff repair. Irrigation pump pressure was recorded at each stage. Hypotensive bradycardic events (HBEs) and postoperative adverse events were monitored.

Results: Overall visual clarity was significantly greater in the EPN group than in the TXA group (mean difference [MD], 0.07; 95% CI, 0.03-0.10; P<0.001). Visualization was superior with EPN during subacromial debridement (MD, 0.08; 95% CI, 0.03-0.13; P=0.001) and rotator cuff repair (MD, 0.03; 95% CI, 0.004-0.06; P=0.02). The EPN group required lower pump pressures and fewer increases in pressure. No significant differences in HBEs or adverse events were observed.

Conclusions: Diluted EPN irrigation improved intraoperative visualization compared with TXA, particularly during the subacromial phases, without increasing perioperative risk. Deep learning provided objective support for its clinical use. Level of evidence: II.