2.1

Participants selection

Utilizing this FDA-approved device, patients undergoing PGCA between January and June 2022 were offered intraprocedural prostate mapping biopsies at time of surgical consent. Twelve participants were consecutively recruited based on availability of the microscope undergoing PGCA by 1 of 2 faculty urologists. Participants were excluded if they received preoperative androgen deprivation therapy.

2.2

Magnetic resonance imaging and magnetic resonance imaging-ultrasound fusion targeted biopsy

As previously described, all MRI were performed with a 3T MRI device with an external phased array coil [ ]. Regions of interest were identified on MRI and scored as per PI-RADSv2 [ ]. MRI to ultrasound fusion targeted biopsy was performed with the Artemis ^TM prostate biopsy system utilizing the Profuse ^TM co-registration software (Eigen, Grass Valley, California) for segmentation and MRI to ultrasound co-registration of potential regions [ ]. All participants underwent targeted and systematic biopsies (4 targeted prostate biopsies taken from the region(s) of interest, in addition to 12 systemic) [ ].

2.3

Partial gland cryosurgical ablation with prostate mapping biopsy technique

All PGCA were performed under general anesthesia in the dorsal lithotomy position. Disease specific treatment plans were designed based upon cognitive or visual estimation of the location of previously segmented regions of interest demonstrating cancer on targeted biopsy. An intended treatment zone was contoured to achieve at least a 10mm intraprostatic margin beyond the tumor segmentation (limited to 3mm beyond any adjacent prostate capsule), while incorporating the location of adjacent systematic cancer-bearing cores [ ]. Between 2 and 4 cryosurgical probes were initially placed under ultrasound guidance. Temperature sensing probes were placed within the freeze zone, at its planned margin and within surrounding critical structures to be preserved. Additionally, a urethral warming catheter was placed over a guidewire for urethral protection during treatment. Two freeze/thaw cycles were carried out for each treatment utilizing the Cryocare CS ^TM system (Endocare, Austin, TX).

Prior to freezing a PB was obtained from the site of anticipated tumor and 2 cores were taken from adjacent 10mm treatment margins based upon MRI predicted tumor volume. During the freeze, biopsy samples were obtained from 3 to 5mm beyond the ice ball. Biopsies were immediately scanned in a SRH microscope which were interpreted by the surgical team intraoperatively. Interpretation of SRH was conducted by M.P.M. and S.S.T. whom have significant experience in the prostate SRH including the development of technique and physician training which included review of over 1000 PCa SRH histologic images [ , ]. Initial experience with physician testing and training for prostate SRH interpretation has been previously described by Mannas et al.[ ] If intraoperative margins returned positive, the freezing was widened to include the positive margin during the second freeze/thaw cycle at the operator’s discretion.

2.4

Creation and interpretation of stimulated Raman histology

Samples were placed in normal saline solution to remain fresh, unstained, and unlabelled prior to scanning in the NIO, a FDA approved stimulated Raman histology microscope (NIO, Invenio Imaging, Santa Clara, CA). As previously described by Mannas et al., PB were placed into a modified, linear NIO slide that compress the samples to a thickness of 225um. Images of 18×1.5mm were then acquired by means of strip-tiling at a fixed imaging depth from the sample surface to create digital image by scanning PBs at a depth of 20 microns using 2 Raman shifts: 2845cm ⁻¹ and 2930cm ⁻¹ [ ]. SRH was created by applying pseudo-coloring to the raw images, as described by Orringer et al, in order to generate a digital image that mimics the appearance of H&E staining [ ]. The SRH microscope produces images during SRH creation which allows for interpretation of the images before completion of scanning. Time to PCa diagnosis and total scan time for each biopsy was recorded. Once the process of creating SRH was completed the PBs were placed into 10% formalin to be processed as per normal pathologic protocols and interpreted by a genitourinary pathologist, utilizing H&E and if necessary, immunohistochemistry.

2.5

Convolutional neural network

Due to the time limitation for intraoperative creation and interpretation of SRH we retrospectively sought to assess if this could be improved by utilizing our previously assessed AI convolutional neural network (CNN) at regular and increased scanning speeds [ ]. Briefly, for the classification of a PB we aggregated the score of all SRH images of that PB by looking at the maximum tumor overall prediction. The previously tested CNN threshold >89% maximum tumor overall prediction for PB malignant classification was utilized as per Mannas et al., otherwise the PB was classified as benign [ ].

2.6

Statistical analysis

All statistical analysis was conducted with SPSS (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0. Armonk, NY: IBM Corp). The sensitivity, specificity, and accuracy of the surgical team and AI interpretation of SRH were compared to the final pathologic result determined by pathologist interpretation of H&E staining with immunohistochemistry, if required. Fisher’s exact test was used to compare accuracy for PCa identification between human and AI interpretation. Atypical small acinar proliferation was excluded from the analysis, as the diagnosis is benign, however, glands may represent a small focus of malignancy.