ECP 2022 Abstract Book

Virchows Archiv (2022) 481 (Suppl 1):S1–S364 13 split into 71 training cases and 20 test cases. A two-step deep-learning algorithm was trained: first all epithelium was detected, and next aber- rant epithelium was distinguished from normal epithelium. Results: The two-step model approach reached an area under the receiver operating curve of 0.946 on slide level. Visual inspection confirms adequate detection of the aberrant epithelium, in concord- ance with morphology and immunohistochemistry. Conclusion: We present a deep-learning algorithm that can successfully detect STIC. Adequate STIC diagnosis is important to better understand the oncogenesis of HGSC, holds prognostic implications for individual patients, and is a prerequisite to safely offer alternative risk reducing surgeries, such as salpingectomy with delayed oophorectomy, currently studied in prospective international trials. We believe an AI model has the potential to aid the pathologist in this challenging diagnosis. Expanding the dataset is expected to aid further development of this model. Funding: Supported by the Dutch Cancer Society. OFP-11-014 Eyeballing and hot-spot counting of ki67 may misguide therapy in invasive breast carcinoma, NST and the quick fix is automated counting S. Sevim*, E. Dicle Serbes, G. Ozdogan, S. Dizbay Sak *Ankara University Medical School, Pathology Department, Turkey Background & objectives: Ki67 evaluation is essential in inva- sive breast carcinomas. This can be a very laborious process for pathologists. In this study, we aimed to compare Ki67 scores by a.)eyeballing, b.)manual-counting (MC), and c.)using an artificial intelligence based automated counting program (AIACP). Methods: For 54 cases, three regions of interests (ROIs) (1 hot- spot and 2 reflecting the first impression), each of 0.2 mm2, were selected on Ki67 (SP6) stained whole digital sections. These ROIs were counted manually and by AIACP (3D-HISTECH, Panoramic P250 Flash3, VPS3.0.2., Quant Center). Blinded to these countings, three independent observers determined Ki67 scores for the same cases, by eyeballing. Results: Conclusion: In this study, our findings of high agreement between AIACP and time-consuming MC shows that a standardized automated Ki67 scoring tool is very beneficial. It should also be stressed that, although interclass agreement between eyeballing and other methods seems acceptable, when Ki67 scores are grouped categorically based on International Ki67 in Breast Cancer Working Group 2021 Consensus, concordance was only moderate. Eyeballing and hotspot-only counting should not be used to determine Ki67 scores, which are critical in determining therapy options. OFP-11-015 AI versus microscope in primary diagnosis of breast biopsies: multi-site clinical reader study A. Vincent-Salomon, A. Nudelman, J. Cyrta, M. Maklakovski, A. Albrecht Shach, G. Sebag, G. Mallel, I. Krasnitsky, T. Feinberg, C. Linhart, M. Vecsler, J. Sandbank* *Maccabi Healthcare Services, Israel Background & objectives: This study aimed to clinically validate the use of an AI-based solution by pathologists for reviewing and reporting breast core needle biopsies as compared with the gold standard practice, review on the microscope. Methods: A two-arm prospective reader study comparing the per- formance of pathologists using an AI-based solution with pathol- ogists using a microscope was performed at two sites (different staining and scanners). Both arms were compared to ground truth (GT) established by consensus of two breast pathologists. Rates of major discrepancies between each arm and GT, as determined by an adjudicating pathologist, were compared. Results: Eight pathologists participated in the study and repor ted on 385 cases (442 HES and 330 H&E slides), each case being reported twice, once in each study arm. Pathologists first reviewed only H&E/HES slides, if requested and available, they were provided with IHCs, while the AI results were on H&E/HES only. The major discrepancy rates of the microscope arm and of the AI arm against GT were 4.42% and 3.12%, respectively, demonstrating 29.4% reduction in major discrepancies. Pathologists with AI demonstrated very high accuracy for the detection of invasive carcinoma with sensitivity and specificity of 100% for both, as well as for DCIS/ADH with sensitivity of 92.4% and specificity of 97.8%. Conclusion: This multi-site reader study reports diagnostic accuracy improvements by pathologists performing diagnosis and reporting with the support of a first read AI solution for breast biopsies. The AI solution performed accurately and generalized well for different staining platforms and different scanners. Thus, AI solutions could be used as significant aiding tools for pathologists in clinical decision-making in routine pathology practice, enhancing the quality and reproducibility of diagnosis. OFP-11-016 Digital score of Ki67 in prostate cancer is associated with high- grade disease and presence of metastasis Q.D. Vu*, L. Mendes, C. D. Brawley, S. Raza, E. Grist, A. Ali, S.S. Vidal, M. Parry, S. Lall, N. B. Atako, M. Richmond, A. Haran, L. Zakka, N. W. Clarke, M. K. B. Parmar, N. D. James, L. C. Brown, D. Berney, G. Attard, N. Rajpoot *University of Warwick, United Kingdom Background & objectives: Ki67 proliferation-index (PI) has been identified as a valuable prognostic marker in prostate cancer. How- ever, manual scoring is laborious and highly subjective. We explore the association of our digital Ki67 score with high-grade prostate cancer disease and presence of metastasis. Methods: Diagnostic paraffin-embedded needle core biopsies were stained for Ki67 using MIB-1 antibody (DAKO, Carpinteria, CA, USA). Manual scoring used the unweighted global assessment method. Digital scores were calculated using our deep learning method. Scores were correlated with Gleason score and metastatic status. Results: Samples from 54 patients randomised to STAMPEDE arm A (2006-2015) were assessed. There were 29 M0, 9 M1 low burden and 6 M1 high burden. Gleason score was centrally assigned with 72% classified as GG5. The correlation between manual and digital scores was calculated to be 0.7562 (p=3.78e-11). The manually assigned Ki67 score was associated with high-grade disease (GG5 and 4) (p = 0.031) and the presence of extra-pelvic metastases (p<0.001), whereas our digital Ki67 score (DigiKiPI) also showed a statistically significant association with high-grade disease and presence of metastasis (p = 0.047 and p = 0.001, respectively) with the added benefits of being objective and reproducible. Conclusion: Ki67 PI is a robust prognostic tool in clinically advanced prostate cancer that can refine patient prognostication. However, it has not been broadly used in clinical routine due to lack of standardisation and high intra and inter observer variations. Automated deep learning based scoring offers a promising tool for better way to objectively and reproducibly quantify Ki67 PI S48