ECP 2023 Abstracts

S272 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Results: Castleman disease (CD) is classified into Unicentric CD (UCD) or Multicentric CD (MCD) depending on lymphadenopathy and systemic symptoms. The infection status of Human herpesvirus type 8 defines the aetiology. Our patient has an iMCD TAFRO-NOS (Not Otherwise Specified), sub-divided in iMCD-idiopathic plasmacytic lymphadenopathy (iMCD-IPL) or iMCD-non IPL, being the latter the one that resem- ble more to our patient. Following this subclassification, our patient still shares clinical and path- ological features between both groups of iMCD-NOS, with significant thrombocytosis, elevated serum IgG levels and indolent clinical course as is expected in IPL, but also pleural effusion, as well as increased vascularity in the architectural pattern that defines it as iMCD non-IPL. Conclusion: Nishikori et al mentioned “there aren´t studies to validate whether or not IPL has distinct clinicopathologic features compared to other iMCD-NOS”, for that reason we present our patienet with iMCD-nonIPL that overlaped clinical traits and matched with the term “oligocentric” or “regional” CD, with multiple enlarged nodes in adja- cent regions and indolent clinical evolution. To conclude the subtlety of the presentation contrast with the large mass, multiple ganglionic areas affected, as well as a large pleural exudate. E-PS-11-011 Artificial intelligence for classifying leukocyte in the peripheral blood: reliable subclassification into 11 types with deep learning algorithm T. Costa*, Y. Usami, M. Iwaya, H. Aguierre, K. Tanaka *Shinshu University Graduate School of Medicine, Science and Tech- nology, Japan Background & objectives: Classifying Leukocyte an important task in the medical routine, which is commonly done with automatic machine’s help, but even the latest machines must be checked by experts. This work aims to improve classification accuracy by using deep learning. Methods: YOLOv7, a new deep learning architecture was used in this study to classify 1779 blood cells images obtained by the CELLAVI- SION DM96 into 11 types of neutrophils (band and segmented), lym- phocyte, monocyte, basophil, eosinophil, lymphocyte variant, meta- myelocyte, myelocyte, thrombocyte, and erythroblast. To detect the abnormal situation more carefully, detailed classification was chosen instead of the typical five types of Leukocytes. Results: We evaluated the simple classification method, where images were classified one time for the neural network with one set of trained weights, and the cascade method, where images were classified in two phases, using two sets of trained weights. Experimental results verified that the cascade method resulted in an accuracy of 96.4% and an F1-score of 98.1%, while the simple method presented an accuracy of 87.4% and an F1-score of 93.28%. Compared with the classification results by CELLA- VISION DM96, an F1-score of 89.71% and an accuracy of 81.33% for the same evaluation database, classified by professionals, we could improve 15.07% of accuracy and 8.39% of F1-score by the cascade approach. Conclusion: We found that our method achieved excellent performance which suggests the possibility of clinical implementation. Even though the classification by the neural network still presented errors, the rate was quite small and was considered irrelevant by the experts. The pro- posed method was considered reliable to the point that the reclassifica- tion by experts of the images after the classification by the machines is no longer necessary, which contributed to reducing the time consumed in the haematology laboratory. E-PS-11-012 Diagnostic utility of digital pathology in the classification and rec- ognition of gastrointestinal non-Hodgkin large B-cell lymphomas – a retrospective-observational study of 15 cases M. Cristian*, M. Aschie, A. Mitroi, A.A. Nicolau, N. Dobrin, I. Poin- areanu, G. Baltatescu *Center for Research and Development of the Morphological and Genetic Studies of Malignant Pathology - CEDMOG, “Ovidius” Uni- versity of Constanta, Romania; Department of Clinical Pathology, “Sf. Apostol Andrei” Emergency County Hospital, Constanta, Romania Background & objectives: Using the example of cases of gastroin- testinal malignant non-Hodgkin large B-cell lymphomas, we want to show that improved visualization and analysis by using digital pathol- ogy contribute to the convergence of these complementary diagnostic modalities of lymphomas. Methods: We have conducted a retrospective study of 50 paraffin- embedded tissues of large B-cell lymphomas diagnosed at the Pathol- ogy Department of the Emergency County Clinical Hospital of Constan ț a and Sacele County Hospital of Brasov, Romania, between 2012 and 2021. All slides were digitized at 40X using the Huron LE120TM 4000XT scanner, Huron Viewer Software, and analysed with the QuPath Software platform. Results: The expression profile of immunohistochemical markers established the diagnosis of non-Hodgkin large B-cell lymphomas, with a high nuclear proliferation marker in most cases. Six patients were female, and nine were male, with ages ranging from 32 to 86 (mean age: 56 years old). Two tumour specimens were sampled from the oropharyngeal region, five were from the stomach, five were from the small intestine, and three were sampled from the large bowel. Cor- roborating these findings, the most often type of lymphoma is DLBCL, NOS and the most often localization in the GI tract is the stomach and the small intestine. Conclusion: The routine histopathological evaluation was assessed by two experienced pathologists with good collaboration, but the automatic identification was more objective and reproducible. In hematopathology, the distinction between various types of NHL using digital microscopy is highly needed to classify images sharing numer- ous morphological features and provide a specifically dedicated tool for accurately diagnosing different types of lymphoma. Also, digital pathology is essential to the continuous quality improvement of medi- cal education. E-PS-11-013 Diffuse large B cell lymphoma CD5-positive arising in an immune deficiency and immune dysregulation setting: A case report and review of the literature M. Cristian*, M. Aschie, R. Baz, A. Stoica, M. Ghinea, M. Deacu, M. Bo ș oteanu, A. Mitroi, N. Dobrin, I.E. Iordache, G. Baltatescu *Center for Research and Development of the Morphological and Genetic Studies of Malignant Pathology - CEDMOG, “Ovidius” Uni- versity of Constanta, Romania; Department of Clinical Pathology, “Sf. Apostol Andrei” Emergency County Hospital, Constanta, Romania Background & objectives: In the era of antiretroviral therapy, lym- phoma is the primary cause of cancer-related death among HIV- infected people. The current report presents the clinicopathological features, diagnostic approach, and clinical outcomes of this HIV-related CD5-positive DLBCL. Methods: We present a case of a 30-year-old male patient with a medi- cal history of HIV-positive serology and antiviral treatment, presenting with diffuse abdominal pain and symptoms related to obstruction or perforation, followed by exploratory laparotomy and surgical resection of the small intestine with other areas of involvement. The surgical specimen was morphologically evaluated and immunohistochemically stained. Results: Histopathological examination revealed a diffuse neoplastic proliferation of large B lymphocytes within the small intestine, lacking features of other defined types of large B cell lymphoma. The CD5-pos- itive DLBCL subtype was diagnosed after immunostaining with twelve monoclonal antibodies (CD3, CD5, CD10, CD20, CD23, CD30, CD68,