External validation of an artificial intelligence model for Gleason grading of prostate cancer on prostatectomy specimens

• DOI: 10.1111/bju.16464
• Publication Date: 2024-07-11T10:11:03Z
• Authors: Bogdana Schmidt; Simon John Christ...; Hriday P. Bhambhvani; Richard E. Fan; Indrani Bhattacharya; Moon Hyung Choi; Christian A. Kunder; Chia‐Sui Kao; John Higgins; Mirabela Rusu; Geoffrey A. Sonn
• Keywords: Grading (engineering); Kappa; Cohen's kappa; Prostate biopsy

Objectives To externally validate the performance of DeepDx Prostate artificial intelligence (AI) algorithm (Deep Bio Inc., Seoul, South Korea) for Gleason grading on whole‐mount prostate histopathology, considering potential variations observed when applying AI models trained biopsy samples to radical prostatectomy (RP) specimens due inherent differences in tissue representation and sample size. Materials Methods The commercially available is an automated system that was previously using 1133 core images validated 700 from two institutions. We assessed algorithm's performance, which outputs patterns (3, 4, or 5), 500 1‐mm 2 tiles created 150 RP a third institution. These were then grouped into grade groups (GGs) comparison with expert pathologist assessments. reference standard International Society Urological Pathology GG as established by experienced uropathologists adjudicate discordant cases. defined main metric agreement standard, Cohen's kappa. Results between pathologists determining GGs at tile level had quadratically weighted kappa 0.94. differentiating cancerous vs non‐cancerous unweighted 0.91. Additionally, classifying 0.89. In distinguishing tissue, achieved sensitivity 0.997 specificity 0.88; ≥2 1 it demonstrated 0.98 0.85. Conclusion excellent cancer identification specimens, despite being entirely different patient population.