Abstract
PURPOSE: Balancing surgical margins and functional outcomes is crucial during radical prostatectomy for prostate cancer. Stimulated Raman histology (SRH) is a novel, real-time imaging technique that provides histologic images of fresh, unprocessed, and unstained tissue within minutes, which can be interpreted by either humans or artificial intelligence. MATERIALS AND METHODS: Twenty-two participants underwent robotic-assisted laparoscopic radical prostatectomy (RALP) with intraoperative SRH surgical bed assessment. Surgeons resected and imaged surgical bed tissue using SRH and adjusted treatment accordingly. An SRH convolutional neural network was developed and tested on 10 consecutive participants. The accuracy, sensitivity, and specificity of the surgical team's interpretation were compared with final histopathologic assessment. RESULTS: A total of 121 SRH periprostatic surgical bed tissue (PSBT) assessments were conducted, an average of 5.5 per participant. The accuracy of the surgical team's SRH interpretation of resected PSBT samples was 98%, with 83% sensitivity and 99% specificity. Intraoperative SRH assessment identified 43% of participants with a pathologic positive surgical margin intraoperatively. PSBT assessment using the convolutional neural network demonstrated no overlap in tumor probability prediction between benign and tumor infiltrated samples, with mean 0.30% (IQR, 0.10%-0.43%) and 26% (IQR, 18%-34%, P < .005), respectively. CONCLUSIONS: SRH demonstrates potential as a valuable tool for real-time intraoperative assessment of surgical margins during RALP. This technique may improve nerve-sparing surgery and facilitate decision-making for further resection, reducing the risk of positive surgical margins and minimizing the risk of recurrence. Further studies with larger cohorts and longer follow-up periods are warranted to confirm the benefits of SRH in RALP.