ECP 2023 Abstracts

S67 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Conclusion: These results suggest the minimal threshold required to detect variants using a single gene test is much lower than the working threshold of 10%. When biopsy tissue is limited, this may have significant implica- tions in the responsible curation of tissue for molecular testing. Funding: Pathological Society of Great Britain and Ireland Undergradu- ate Elective Bursary PS-02-009 Deep learning for multi-class cell detection in H&E-stained slides of diffuse gastric cancer R. Lomans*, J. van der Laak, I.D. Nagtegaal, F. Ciompi, R.S. van der Post *Radboud university medical centre, The Netherlands Background & objectives: Diffuse gastric cancer (DGC) is character- ized by poorly cohesive cells which are difficult to detect. We propose the first deep learning model to detect classical signet ring cells (SRCs), atypical SRCs, and poorly differentiated cells in H&E-stained slides of DGC. Methods: We collected slides from 9 patients with hereditary DGC, resulting in 105 and 3 whole-slide images (WSIs) of gastric resections and biopsies, respectively. The three target cell types were annotated, resulting in 24,695 cell-level annotations. We trained a deep learn- ing model with the Faster-RCNN architecture using 99 WSIs in the development set. Results: The algorithm was tested on 9 WSIs in the independent valida- tion set. Model predictions were counted as correct if they were within a 15-micron radius from the expert reference annotations. For evaluation, we split the detection task into two components: class-independent cell localiza- tion (recognition of any tumour cell type) and cell-type classification (catego- rizing localized cells as the correct types). We found (average) F1 scores of 0.69 and 0.93 for the localization and classification tasks, respectively. Thus, we observe that the algorithmdoes not generallymisclassify cells, but rather, the errors mainly arise frommissing cells or false positive predictions of cells that do not belong to the three target classes. Conclusion: Future work will focus on improving the cell localization performance of the algorithm. Cell localization of the three target classes will be an important task in a clinical application of our model, in which it could be used to improve the detection of DGC lesions among large sets of slides. Moreover, the algorithm will allow for quantitative assessment of DGC patterns, potentially giving new insights in specific morphological features of DGC such as patterns of spatial cell distributions. This research was supported by an unrestricted grant of Stichting Hanarth Fonds, The Netherlands PS-02-010 IMSeg: a clinically deployable tool for the detection of intestinal metaplasia in gallbladder using digital pathology slides and deep learning J. Massonnet*, K. Egervari, M. Kreutzfeldt, D. Merkler, A. Janowczyk *Department of Pathology, Division of Clinical Pathology Geneva University and University Hospitals, France Background & objectives: Early detection of gallbladder intestinal metaplasia (IM) is critical for improving patient outcomes, as IM can ultimately progress to dysplasia and cancer. To ease pathologist’s exhaustive high-magnification screening, a deep learning tool identi- fying IM on H&E-stained slides, IMSeg, was developed. Methods: Emergence of goblet cells in the epithelium is a defining feature of IM. Consequently, IMSeg takes a 2-step filtering approach: 1) segmentation (UNet) of epithelial regions at 5x magnification, 2) subsequent segmentation of masked regions for goblet cell detection (UNet) at 10x. Patient-level diagnostic accuracy, and goblet cell level F1 scores were computed using n=50 patients (10 IM-positive). Results: Calibrated using 80/20 training/testing split, patient-level diagnostic accuracy of 100% was observed. At the goblet cell level, an F1-score of 0.95 was observed, with successful detection determined by a Dice coefficient ≥ 0.85. IMseg’s output directs pathologist atten- tion directly to IM regions, eliminating the need to manually review the entire slide, and resulting in a 50% (2.5 minutes) reduction in patient diagnostic time. This translates to an anticipated yearly time saving of 40 hours in our institution. Furthermore, detection of IM by IMSeg will automatically prompt the preparation of additional tissue blocks, necessary for ruling out dysplasia in such cases, ultimately reducing patient case turn-around time by approximately one day. Conclusion: Automated detection of IMprovides a powerful, highly accu- rate, tool for pathologists. IMseg was designed to be seamlessly integrated into our hospital’s primary digital diagnostic workflow, helping streamline the diagnostic process and improve lab efficiency. Future work will focus on performing that integration while evaluating the accommodation of pathologists using IMSeg in their daily practice, as well as maintenance to ensure its performance remains sufficient for clinical usage over time. PS-02-011 IBAPAT project: the unified digital pathology laboratory of the Balearic Islands G. Matheu Capó*, D. Laguna Macarrilla, A. Forteza Valadés, F.A. Miguel Gayá, C. Fernández Palomeque *Servei d’Anatomia Patològica. Hospital Universitari Son Espases, Spain Background & objectives: The Balearic Islands (BI) have six Pathol- ogy Departments spread over three islands, with variable technology and specialization. Digital pathology can facilitate the creation of a single pathology laboratory to minimize accessibility and care equity problems owing to insularity. Methods: The IBAPAT project is based on three axes that are being developed sequentially: 1. the implementation of a Laboratory Informa- tion System (LIS) for the six Departments of Pathology in the BI; 2. Whole-slide imaging (WSI) of all histological preparations for routine diagnosis; 3. The centralization of laboratory processes in a single digitized and highly productive laboratory for all the BI. Results: An LIS has been acquired; it is a single database for all pathology departments, with a traceability system adapted to digital pathology, allowing networking and management of a central labora- tory. Scanners covering all histological activities (single laboratory model), storage hardware, image viewer software, and basic algorithms for diagnosis in digital pathology have been developed. The central lab- oratory building is in the construction project phase, from tissue pro- cessing to stained slides and their scanning, and other ancillary studies (FISH, PCR, NGS, etc.) will be performed in this laboratory. Patholo- gists maintain their work activities in their pathology departments. Conclusion: Taking into account the insular geographical context and the need to give patients equal and universal access to health care and technology, the IBAPAT project pursues the modernization of IB Pathology departments by improving information systems, theWSI, and centralization of the labora- tory, improving the global efficiency of the diagnostic process in pathology, and improving patient care in terms of equity, safety, and quality of diagnosis. PS-02-012 Predicting cell-of-origin for diffuse large B-cell lymphoma patients (DLBCL) using explainable feature-based model P. Lin, N. Shaikh, P. Porwal, S. Jayachandran, Q. Gu, X. Li, K. Korski*, Y. Nie *Department of Personalized Healthcare, Data, Analytics and Imaging group, Genentech, Inc., USA Background & objectives: Determining cell-of-origin (COO) sub- types in DLBCL has been only possible by genomic testing or multiple