ECP 2023 Abstracts

S138 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Background & objectives: The increasing prevalence of renal cancers necessitates accurate detection and subtyping. However, diverse histo- logic spectrums make this task challenging. Recently, AI has demon- strated potential in histologic image analysis but has been underutilized in renal cancer due to dataset limitations. Methods: To address this, we aim to create a deep learning model for renal cell carcinoma (RCC) diagnosis by utilizing whole slide images (WSIs) sourced from multiple hospitals, which could enhance the accu- racy and efficiency. A total of 1,639 WSIs from two university hospi- tals, Seoul and Uijeongbu St. Mary’s hospital, were used to compare ten convolutional neural network models. Results: Accuracy, area under the curve (AUC), and F1 score of resnet18, resnet34, resnet50, resnet101, densenet, mobilenet, vgg, efficientnet, inception, squeezenet were compared after normalization and augmentation for efficient learning. Images were divided into train- ing, validation, and test sets for model development and evaluation. The Efficientnet model demonstrated the highest performance, with an accuracy of 96.82%, AUC of 98.67%, and F1 score of 97.80%. Conclusion: Our study indicates that employing various deep learning models can enhance the accuracy and efficiency of RCC diagnosis using WSIs. These promising results in renal neoplasms detection suggest potential for future advancements in RCC diagnosis and treatment. How- ever, further external validation with larger datasets from multiple insti- tutes and ethnicity is needed to improve performance and generalizability. PS-26-006 Routine use of an artificial intelligence solution for primary diag- nosis of prostate biopsies in clinical practice R. Dhir*, R. Abou Shaar, G.M. Quiroga-Garza, K. Takeda, D. Hart- man, M. O’Leary, R. Ziv, M. Grinwald, M. Vecsler *University of Pittsburgh Medical Center, USA Background & objectives: We present the analysis of a clinically deployed artificial intelligence (AI) decision support solution for pros- tate biopsies primary diagnosis utilized as first read within a digital pathology workflow. Methods: The AI solution was previously validated in the lab on an independent cohort. Four pathologists underwent training and used the solution for prospective primary diagnosis of consecutive prostate core needle biopsies, reporting on 122 cases (509 parts, 770 H&E slides). Results: The AI solution demonstrated high performance when pre- classifying parts with the likelihood to be benign or malignant, with AUC = 0.99 (95%CI: 0.985, 0.997), NPV = 99.9% (95%CI: 0.966, 0.999) (163/164) and PPV = 97.6% (95%CI: 0.948, 0.991) (244/250), respectively. 19% of parts have been classified as suspicious by AI. The AI performance in Gleason group grading was high, at 90.6% of full agreement or one group difference. User feedback survey, showed high satisfaction marks for the AI solution, particularly for the Glea- son scoring (95%), PNI detection (90%), and tissue and tumour length automated measurement (95%). Pathologists felt there is potential to increase diagnostic efficiency by using the AI tool. Conclusion: We report here the successful implementation of a multi- feature AI solution that automatically imparts clinically relevant diag- nostic parameters regarding prostate cancer, grading, measurements, and other pathologic features. The solution demonstrated its ability to accurately detect cancer and contribute to diagnostic quality. Thus, the AI solution could be used as a significant aiding tool for pathologists in clinical decision-making in routine pathology practice. PS-26-007 Characterisation of embryonic-type neuroectodermal tumour and embryonic-type neuroectodermal elements adopting a broad immunohistochemical panel L. Di Sciascio*, F. Ambrosi, T. Franceschini, F. Giunchi, G. Di Filippo, E. Franchini, F. Massari, V. Mollica, V. Tateo, F. Mineo Bianchi, M. Colecchia, A.M. Acosta, J. Lobo, M. Fiorentino, C. Ricci *School of Anatomic Pathology, Department of Biomedical and Neu- romotor Sciences, University of Bologna, Italy Background & objectives: Embryonic-type neuroectodermal tumour (ETNT) is an aggressive somatic-type malignancy of the testis. For its rarity, there’re limited data on its immunohistochemical features. Herein, we tested a series of ETNT with a broad panel to clarify the immunohistochemistry aiding the diagnosis. Methods: Twelve cases including 4 ETNT, 4 teratomas of the testis with embryonic-type neuroectodermal elements (ETNE), and 4 imma- ture teratomas of the ovary and/or the central nervous system were collected. The cases were tested with a broad immunohistochemical panel including SOX2, NF, INI-1/SMARCB1, SMARCA4/BRG1, S-100, SOX10, NeuN, WT-1, CD99, GFAP, Synaptophysin, Chro- mogranin, and CK AE1/AE3, and adopting a previously-described scoring system. Results: All cases were reviewed and classified according to the spe- cific WHO classification systems. ETNT displayed a wide range of histological patterns (atypical small round/epithelioid/spindle cells, high mitotic count, solid sheets/fascicular arrangement/multilayered rosettes/anastomosing neural tubules, and necrosis) often merged with mature glial/neural components. SOX2 (mv: 6.4, r: 0-9) and cytoplasmatic WT-1 (mv: 6.3, r: 3-9) were the most frequent and intense stains for the immature neuroepithelial components of all the selected histological entities. Cytoplasmatic WT-1 showed the best values also for the mature glial/neural components. All the other stains were completely negative or focally/weakly positive in the immature neuroepithelial components, with variable results in the mature glial/ neural components. Conclusion: Our study suggests that SOX2 and cytoplasmatic WT-1 are useful diagnostic tools for the identification of immature neuroepi- thelial components in germ cell tumours of the testis, so aiding the diagnosis of ETNT and quantitative distinction from teratoma with ETNE. In our study, SOX2, a pivotal factor for the development of neuroepithelium, did not stain mature glial/neural components and performed better than cytoplasmatic WT-1 at helping estimate of the amount of immature neuroepithelial components. PS-26-008 Functional atlas of prostate cancer mesenchyme: a translational approach to untangle the stromal molecular landscape in prostate cancer initiation, progression, and metastatization via single-cell profiling G.N. Fanelli*, H. Pakula, M. Omar, R. Carelli, F. Pederzoli, T. Pan- nellini, S. Rodrigues, C. Fidalgo-Ribeiro, P.V. Nuzzo, D.S. Rickman, B.D. Robinson, A.G. Naccarato, C. Scatena, L. Marchionni, M. Loda *University of Pisa, Italy Background & objectives: Prostate cancer (PCa) has divergent clini- cal behaviour that molecular alteration in epithelial cancer cells can only partially justify. Indeed, tumour stroma can deeply influence it. Using scRNA-seq we explored mesenchymal cells’ expression pro- grams to untangle stromal impact on PCa carcinogenesis. Methods: We used four PCa mouse models representing different disease stages: TMPRSS2-ERG (T-ERG) model for PCa initiation; Nkx3.1creERT2;Ptenf/f (NP) model for intraepithelial neoplasm; Tg(ARR2/Pbsn-MYC)7Key (Hi-MYC) for early invasive adeno- carcinoma and Pb-Cre4+/-;Ptenf/f;Rb1f/f;LSL-MYCN+/+ (PRN) for advanced adenocarcinoma with neuroendocrine features. Using scRNA-seq we compared their mesenchymal cells’ transcriptional program to their wild-type counterparts, and to primary and meta- static human samples with comparable genotypes.