ECP 2023 Abstracts

S234 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Clinical benefit was estimated via comparison of standard and HPy- loriDet-aided diagnostic times. Results: On extracted patches, HPyloriDet achieved an accuracy of 95%, alongside 92% sensitivity and 95% specificity. The positive pre- dictive value was 65% while the negative predictive value reached 99%. For slides confirmed as positive, HPyloriDet reduced Helicobacter pylori diagnosis time by up to 80% (3 minutes) without producing false negatives. For negative slides, HPyloriDet showed a 25% (1 min- ute) improvement. The benefit for negative slides was more modest as they still required careful pathologist review to avoid false negative diagnoses, whereas for positive slides, correct detection alone appears sufficient. HPyloriDet is thus estimated to provide a yearly time sav- ings of 100 hours based on a typical yearly workload of 5000 slides. Conclusion: Standard HP detection on large IHC slides is time-con- suming and our computer-aided tool HPyloriDet demonstrates prelimi- nary evidence that this burden can be ameliorated. Preliminary find- ings indicate potential time-saving benefits in clinical settings, without loss of diagnostic accuracy. Future work will involve validating the time-efficiency when integrating HPyloriDet into the clinical work- flow, monitoring its performance, and collecting additional data during deployment for retraining to further improve HPyloriDet’s accuracy. E-PS-08-009 Immunohistochemical evidence of mTOR C1 and C2 pathway in diffuse idiopathic pulmonary neuroendocrine cell hyperplasia (DIPNECH) - a digital pathology analysis L. Comerford*, M. O’Callaghan, M. Alquier, J. McCormack, E. Lynn, O. O’Carroll, D. Murphy, R. Crowley, D. O’Toole, C. McCarthy, A. Fabre *Histopathology department, St. Vincent’s University Hospital, Dublin, Ireland Background & objectives: Background: DIPNECHmay represent a pre- cursor to neuroendocrine tumours and is confined to the bronchial epithe- lium. DIPNECH is a rare disease affecting essentially women and can pre- sent with nodules on HRCT, some in association with carcinoid tumours. Methods: Rare reports of altered signalling via the mammalian target of rapamycin (mTOR) pathway are described in neuroendocrine cell diseases. Objective: To quantify tissue expression of specific mTOR pathway downstream proteins RPS6 and 4EBP1 on DIPNECH and control lung tissue sections using immunohistochemistry and evaluation by digital pathology. Tissue analysed from 15 patients with a pathological diagnosis of DIPNECH, aged 26-80. Five symptomatic (cough, dyspnoea); two MEN syndrome, 4 previous breast cancer, 11 carcinoids, 2 multiple tumorlets. 7 control lung tissue without DIPNECH from lobectomies sampled away from lesions included. Slides stained with Chromogranin, RPS6 and 4EBP1, ER and PR anti- bodies using Dako Autostainer48 and uploaded using Leica AperioAT2 scanner. ImageScope Pathology Slide Viewer was used to facilitate annotation of bronchioles. Image analysis of annotated areas was used to calculate (1) number of chromogranin positive cells per millimetre squared of bronchiole, (2) % chromogranin cells per bronchiole, (3) % 4EBP-1 cells per bronchiole and (4) % RPS6 cells per bronchiole, ER and PR expression. Results: 4EBP1 is ubiquitously expressed in resident pulmonary epithe- lial cells and expression was observed in all cases with neuroendocrine cell hyperplasia and tumorlets. RPS6 expression varies within resident cells and is expressed in the majority of neuroendocrine cells but not as diffusely as 4EBP1. ER and PR were negative in all neuroendocrine cells. Conclusion: Our results provide supporting evidence for the role of mTOR signalling in DIPNECH, with mTORC and C2 pathway protein expression in DIPNECH and might support the role of mTOR inhibi- tors in treatment of symptomatic patients. E-PS-08-010 Validation of a quantitative image analysis algorithm for Ki67 index in breast cancer and neuroendocrine tumour M. Cossutta*, J. Bellocq, C. Egele, A. Papine, E. B. Tatarinova, K. Socha, M. Soussaline, C. Homsy, F. Soussaline *IMSTAR Dx, France Background & objectives: Ki67 index evaluation in Breast Carcinoma (BC) and Neuroendocrine Tumour (NET) depends on the quality of immunohistochemistry (IHC), its interpretation by the pathologist and the spatial intratumoral heterogeneity. Our objective was to design an adapted Quantative Image Analysis (QIA) algorithm. Methods: The evaluation used 121 slides providing from 2 external proficiency testing schemes (2021, n=55 slides; 2022, n=66 slides) conducted by the french interlaboratory comparison organization (AFAQAP), and comprising 2 different BCs and 2 different NETs. All indexes were assessed by 2 independent methods: visually by 2 expert pathologists, and by QIA using the IMSTAR PathoScan Tumour- Marker Ki67 algorithm. Results: Pathologists identified 4 classes of Ki67 index for each BC and NET depending on IHC technical quality (optimal, good, border- line, insufficient): 46 Ki67 IHC slides received the ”optimal technique” appreciation by the experts for each BC and NET (2021, n= 20; 2022, n= 26), with Ki67 indexes of 12-15% (2021) and 5-10% (2022) for BCs, 3-5% (2021) and 4-5% (2022) for NETs. Ki67-QIA algorithm performed an accurate evaluation of Ki67 index with a concordance (±1% outside the index classes) with the experts of 89% for BCs and 96% for NETs. This 89% for BCs results from the mixing of an het- erogeneous and an homogeneous tumour (80% and 95% concordance, respectively). Conclusion: The QIA solution was efficient to evaluate Ki67 index on technically optimal IHC slides despite multiple laboratories IHC techniques being applied, highlighting the need for quality in IHC to obtain robust digital evaluations. The objective quantification of intra-tumoral heterogeneity opens an additional challenge. Ki67-QIA algorithm allows to evaluate Ki67 index on a large number of cancer cells and to visualize and quantify Ki67 index spatial variations in tumours, enabling to identify tumour zones under or over the clinical utility thresholds. E-PS-08-011 A complete clinically applicable lung cancer diagnostic platform based on histopathological artificial intelligence J. Da*, N. Che, D. Zhao, Y. Zhao, S. Wang *Dept. of Pathology, China Background & objectives: Lung cancer is a leading cause of cancer- related deaths worldwide. We aim to establish an artificial intelligence (AI)-powered platform for lung cancer detection and subtype classifica- tion, which is crucial for precise treatment of lung cancer. Methods: We collected 1,115 lung slides and digitized them into whole-slide images (WSIs) and randomly divided the WSIs into train- ing, validation, and test sets. Additional WSIs for non-mucinous adeno- carcinoma subtype classification were also collected. Using DeepLab v3 image segmentation model, we established pixel-level lung cancer detection and subtype classification models. Data augmentation tech- niques were applied for model robustness. Results: The deep learning model achieved clinical-grade performance in lung cancer detection and main subtype classification. Based on the cancer detection model (AUC: 0.970, sensitivity: 94.1%, speci- ficity: 94.6%), the main subtype classification model for squamous cell carcinoma, adenocarcinoma, and small cell carcinoma reached a sensitivity/specificity of: (test set) 88.6%/81.8%, 89.7%/83.1%, and 83.3%/94.3%; (surgical specimens) 87.6%/79.7%, 89.5%/80.9%, and 82.4%/75.0%; (biopsy specimens) 92.9%/85.8%, 91.7%/86.2%, and