ECP 2023 Abstracts
S103 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Research Grant (BO/00723/22 to BR). The funding body did not play roles in the design of the study and collection, analysis, and interpreta- tion of data and in writing the manuscript. PS-17 | Poster Session Digestive Diseases Pathology - GI PS-17-001 A weakly supervised learning approach based on multiple instance- learning and self-supervised contrastive learning for gastric cancer classification using whole slide images. A., S. Ali*, T.P. Theodore Armand, H. Kim *Inje University, Republic of Korea Background & objectives: The whole slide image classification is a challenging task in digital histopathology because they are big in size and have high resolution. It is very difficult and time-consuming to annotate WSI. We used multiple instance learning (MIL) for gastric cancer. Methods: To deal with the WSI classification of gastric cancer, we applied dual-stream MIL. Since WSI contains only a small portion with disease-positive instances resulting in data imbalance between positive and negative bags, therefore, to address this problem, we introduced self-supervised contrastive learning. It extracts useful rep- resentations for MIL and addresses the issue of prohibitive memory costs for large bags. Results: We collected gastric cancer data from five different hospitals in South Korea. The dataset consists of 220 training and 54 testing WSI images which produced 1.2 million patches of size 224 x 224 at 10x magnification. Each slide contains pixel-level annotations of the tumour regions and is weakly annotated. However, we consider slide- level annotations. We trained the backbone model resnet18 on 500 epochs. Moreover, for contrastive learning, the SimCLR model was trained on 100 epochs. We calculated the accuracy and area under the curve (AUC) of our model for the classification of poorly differentiated (PD) and well-moderately differentiated (WMD) classes. The model achieved 73% accuracy and 80.5% AUC. Conclusion: In this research study, we introduced a weakly supervised- based dual-stream MIL approach for the classification of WSI. We inte- grated self-supervised contrastive learning with MIL. In DS-MIL, the highest-scored instance is determined by the first stream using standard max pooling, while by determining each instance’s distance from the highest-scored instance, the second stream calculates an attention score for each instance. We think that our approach represents a significant advancement in the diagnosis of gastric cancer using WSI. This research was supported by the MSIT(Ministry of Science ICT), Korea, under the National Program for Excellence in SW, supervised by the IITP(Institute of Information & Communications Technology Planning & Evaluation) in 2022 (2022-0-01091, 1711175863). PS-17-002 Does the morphology of Helicobacter pylori matters for the histo- pathology in gastric endoscopic biopsies? Y. Adali*, Ö. Ertener, H. Beşeren, G. Yıldız *Izmir University of Economics Faculty of Medicine, Turkey Background & objectives: The histopathological changes in the presence of Helicobacter pylori (H. pylori) in the stomach are well defined. In this study, it is aimed to evaluate the relationship between the main morphological types of H. Pylori and histopathological find- ings observed in gastric endoscopic biopsies. Methods: The study included 791 gastric endoscopic biopsies evalu- ated by a single pathologist over a 2-year period. In addition to the mor- phological findings re-evaluated according to the updated Sydney clas- sification by two separate pathologists, the morphological appearances of H. Pylori was categorized as bacillus, coccoid, bacillus+coccoid forms. Afterwards, H. Pylori morphology, demographic information and histomorphological findings were compared. Results: Active chronic gastritis was observed in 442(55.9%) cases and chronic gastritis was observed in 338(42.7%) cases. The presence of H. Pylori was observed in 604(76.4%) cases. The morphology was bacil- lus in 456(57.7%) cases, coccoid in 70(8.8%), and bacillus+coccoid in 78(9.9%). In cases with active chronic gastritis, the most statisti- cally significant morphology is bacillus, then bacillus+coccoid, and lastly coccoid form (p=0.000). In cases with chronic gastri- tis, bacillus+coccoid, then bacillus and then coccoid morphologies are observed most (p=0.015). While no statistical significance was observed between intestinal metaplasia and dysplasia and morphology, the presence of lymphoid aggregates was mostly observed in bacillus, then bacillus+coccoid and last coccoid forms (p values 0.317, 0.402, 0.000, respectively). Conclusion: The relationship between the bacillus form and active chronic gastritis has been shown, as expected from routine pathology practice. However, the change in this morphological order, which is expected in chronic gastritis, indicates that the history of anti-gastritis and anti-acid drugs should be carefully examined. In addition, it is thought that no statistically significant relationship was found between H. Pylori morphology compared to intestinal metaplasia and dysplasia due to the low number of cases with these histopathological findings. PS-17-003 Clinicopathological spectrum of primary anorectal melanoma: a single centre experience of 36 resected cases T. Agrawal*, M. Bal, K. Deodhar, M. Ramadwar, A. Saklani, S. Yadav, R. Kumar *Tata Memorial Hospital, India Background & objectives: Anorectal melanoma is an uncommon and aggressive malignancy with poor median survival rates. In this study we aimed to evaluate the clinical and histopathological parameters and to correlate these clinic-pathologic parameters with prognosis. Methods: Operated cases of anorectal melanoma from 2013 to 2019 were included. Clinical and treatment details were obtained from the electronic medical record. Pathological parameters evaluated included histological type, tumour depth, level of anorectal wall invasion, vertical and radial growth phase, mitotic rate, surface ulceration, lymphovascular invasion, perineural invasion, microscopic satellitosis, tumour infiltrating lymphocytes, and lymph node status. Results: A total of 36 cases were assessed. The median age was 53.5 years. Tumour epicentre was anorectal junction(24), rectum(7) and anal canal(5). Median tumour dimension was 40 mm(7-150mm) and histologic depth of 13.5 mm(1-23mm). Histomorphology was epithelioid (44.5%, n=16), spindled (11%, n=4) and mixed (44.5%, n=16). Invasion into muscularis propria and beyond was seen in 22 patients (61%). Radial growth phase was seen in 50%(n=17/34), pagetoid spread(26%, n=9/34), junctional activity(35%, n=11/31), lymphovascular invasion(22%, n=8), perineural invasion(11%, n=4), and microsatellitosis (17%, n=6). Amelanotic cases were 19%(n=7). Lymph node metastasis was seen in 55%(17/31), whereas 85%(22/26) developed distant metastasis. Follow-up data was available for 22 patients and 7(32%) were alive (median follow-up: 13 months). Conclusion: Anorectal melanoma is an aggressive neoplasm and patients present at an advanced stage. These patients have a higher frequency of lymph node metastasis and distant metastasis, and an overall poor survival rate. PS-17-004 TempO-Seq Detects Fusobacterium nucleatum in formalin- fixed paraffin-embedded patient tissue from a bowel screening population S.S.F. Al-Badran*, S. McSorley, M. Johnstone, J. Hay, J. Jawny, N. Maka, D. Andersen, G. Lynch, J. Edwards *School of Cancer Sciences, University of Glasgow, United Kingdom
Made with FlippingBook
RkJQdWJsaXNoZXIy Mzg2Mjgy