Enhancing the molecular assessment of MRI targeted prostate biopsies utilizing near real time stimulated Raman histology and artificial intelligence.

Abstract

384 Background: Stimulated Raman histology (SRH) produces rapid, label-free optical sections of fresh tissue that can be interpreted within minutes by artificial intelligence (AI). Standard tissue processing hinders molecular assessment of prostate cancer (PCa) necessary for advancement of precision medicine. Our objectives were to prospectively validate the NYU SRH-AI algorithm in an MRI targeted biopsy cohort and to test whether SRH-guided tissue banking improves downstream tissue for precision oncology in a real-world biopsy workflow. Methods: 200 men with a PI-RADS 3-5 (n=256 regions of interest (ROI)) undergoing a transperineal targeted prostate biopsy (TB) were prospectively enrolled in an IRB approved study. The TB were kept fresh before scanning with the NIO SRH microscope (Invenio Imaging Inc, Santa Clara, CA) using two Raman spectra: 2845cm -1 and 2930cm -1 . Both spectra are required for human interpretation but AI interpretation can use rapid scanning parameters utilizing only 2845cm -1 . The NYU SRH-AI algorithm was incorporated into the SRH microscope to provide a near real-time PCa identification and area quantification. After SRH imaging, TB were placed in liquid nitrogen for tissue biobanking or formalin for routine pathologic processing and ground truth diagnosis. Of the 200 men, 163 participants with 200 ROIs biopsies were assessed for tissue banking. Seventeen samples were selected for DNAseq, and the selection was stratified to reflect the cohort distribution across PCa grade groups, SRH-AI estimated cancer area, core length, and the number of trimming iterations. DNA were extracted from cryobanked tissues, sequenced, and then compared against standard workflows utilizing formalin fixed paraffin embedded (FFPE) tissues. Results: The full-scan model achieved a concordance index of 0.935 for PCa identification in 5 mins. The rapid-scan model achieved a concordance index of 0.930 for PCa identification in 2.5 mins. Among 200 ROIs, SRH-guided triage significantly enriched tumor content in banked cores: median SRH suspected tumor area of TB-banked with PCa 71% (IQR 58–81; N=65) vs non-banked ROI with PCa 11% (IQR 5–20; N=51), and non-banked benign ROI 5% (IQR 3–8; N=84). DNAseq showed SRH-AI selected cryopreserved samples demonstrated tumor fraction enrichment of 0.61 (0.49-0.73), compared to conventional FFPE sampling 0.48 (0.22-0.64), (p<0.001). In addition, DNA seq copy number noise was significantly reduced in SRH/AI samples 0.013 (0.012-0.013) compared to FFPE 0.11 (0.085-0.16), (p<0.001). Conclusions: SRH-AI identified PCa in diagnostic biopsies in 2.5 mins guiding biobanking of tumor enriched tissues, yielding higher-quality DNA for molecular analyses.