ECP 2023 Abstracts

S241 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 *School of Anatomic Pathology, Department of Biomedical and Neu- romotor Sciences, University of Bologna, Italy Background & objectives: Digital pathology has the potential to transform the practice of diagnostic pathology but is still an emerg- ing field. This study evaluates interobserver agreement in the digital evaluation of serrated lesions/polyps on a computerized tool in a group of 21 pathologists. Methods: A set of 50 routinely digitized serrated lesions/polyp lesions were reviewed by 21 pathologists belonging to 12 Italian Pathology Departments. For each case, an hematoxylin &eosin slide was evalu- ated according to the morphologic criteria of the WHO classification of Digestive System Tumour, and all responses were recorded. The results were statistically analysed using the Fleiss’K to assess concord- ance of diagnoses. Results: The cohort included 50 patients: 23 males and 27 females, with a mean age of 68 years (range 50-89). First, the agreement in the diagnosis of serrated lesions and polyps resulted fair (0.45 Fleiss’ k) among all participants. Interestingly, the agreement increased to moderate among all pathologists from the Bologna area (0.61 Fleiss’ k). In particular, the kappa value resulted moderate (0.58 Fleiss’ k) and substantial (0.53 Fleiss’ k) in the two Bologna pathology insti- tutes, respectively. In contrast, the agreement among the remaining 12 pathologists from other pathology departments was fair (0.42 Fleiss’ k). Conclusion: Agreement and reliability in diagnoses of serrated polyps/ lesions on digital pathology were found to be higher among patholo- gists belonging to the same institution. The present results demonstrate the importance of working with shared criteria to be more reproducible in the diagnosis of controversial lesions. E-PS-08-037 End-to-end pipeline for automatic grading of IHC biomarkers C. Simmat*, N. Nerrienet, R. Peyret, M. Petit, S. Touioui, A. Chia- verini, M. Sockeel, B. Jean Jacques, S. Sockeel, E. Lanteri *Primaa, France Background & objectives: Immuno-histochemistry (IHC) is a stain- ing process that highlights prognostic and predictive biomarkers. This process is tedious and time consuming. It has also proven to hold high inter-observer variability. Artificial intelligence (AI)-based systems could assist pathologists in clinical diagnosis. Methods: Existing tools are only semi-automatic and require pathol- ogists to select invasive carcinoma (IC) regions on IHC slides manu- ally or by registration. We leverage a CycleGAN-based data aug- mentation strategy to train a model that identifies cancerous regions on IHC slides. In those regions, we apply our automated scoring method for IHC biomarkers including Ki-67 and ER/PR based on AI methods. Results: We show that our method consistently and precisely locates IC regions in IHC WSIs directly by comparison with pathologists ground truth annotations on a set of WSIs. We also demonstrate the robustness of our automatic IHC grading tool with accuracy and efficiency metrics at slide-level on a multi-centric dataset. Conclusion: In this work, we introduce an end-to-end automatic pipe- line capable of * identifying IC regions directly on IHC WSIs with no need of human intervention or registration step and * performing the grading process quickly and accurately. Therefore, our method has the potential to streamline pathology laboratories workflow significantly. E-PS-08-038 Spatial correlation between tumour microenvironment and in vivo perfusion CT parameters H. Smits*, E. Bennink, L. Ruiter, G. Breimer, S.M. Willems, J.W. Dankbaar, M. Philippens *University Medical Center Utrecht, The Netherlands Background & objectives: In this study, we investigated the spa- tial correlation between immunohistochemistry (IHC) biomarkers Ki-67 (proliferation), HIF-1α (hypoxia), and CD45 (immune cells), and in vivo dynamic contrast enhanced (DCE-)CT parameters to test the feasibility of characterizing the tumour microenvironment with DCE-CT. Methods: 56 whole mount tumour slides of 15 laryngeal and hypopharyngeal carcinomas were immunohistochemically stained and digitized. Heatmaps of biomarker positivity were created, registered to the in vivo maps of Ktrans (transfer constant), Ve (extravascular and extracellular space), and Vi (intravascular space), and downsampled to 3x3x3mm3. Pearson’s correlation coefficient r between IHC biomarkers and DCE-CT parameters was determined for each tumour. Results: After a Bonferroni correction for multiple testing, a sig- nificant negative correlation was found between Ki-67 and Ve (mean r=-0.15, range -0.02 to -0.32, t-test P<.001). This indicates that DCE- CT gives relevant information about the microenvironment, as it is to be expected that cell proliferation and extracellular space are nega- tively correlated. Additionally, we found a significant correlation between Ki-67 and HIF-1α (mean r=0.27, range 0.06 to 0.54, P<.001). This might be attrib- uted to continuing cell proliferation that creates temporary or mild reduc- tion of oxygen supply within a tumour, causing upregulation of HIF-1α. No significant correlations were found between any perfusion param- eters and CD45 or HIF-1α. Conclusion: This unique study shows the technical feasibility of cal- culating the spatial correlation between pathological biomarker heat- maps and in vivo imaging. In vivo contrast distribution gives relevant information about tumour microenvironment characteristics. Funding: The material used in this paper was collected in a study funded by the Dutch Cancer Society (KWF) Research Fund, project number 2011-5152. Additional funding for analysis was provided by the same society, project number 2017-10978. E-PS-08-039 Imaging analysis of nuclear features of high-grade nonmuscle inva- sive urothelial carcinoma - diagnosis, staging and relapse V. Sousa*, R. Rodrigues, M.R. Silva, M. Viseu, C. Eliseu, T. Ferreira, A.F. Ladeirinha, A. Alarcão, A.L. Alves, G. Fontinha, R. Almeida *IAP-PM - Universidade de Coimbra, CHUC, Portugal Background & objectives: Bladder cancer remains the sixth most prevalent cancer in man. Whole-Slide-Imaging(WSI) and image analy- sis techniques have demonstrated great utility in Pathology. Associa- tions between nuclear morphometric features and histopathological type and recurrence rate were investigated on high-grade non-muscle- invasive urothelial bladder carcinoma(NMIBC). Methods: We selected 55 patients diagnosed with NMIBC at Centro Hospitalar Universitário de Coimbra between 2014 and 2017. From each patient, a representative HE slide of the neoplasm was selected for scanning and application of an image analysis algorithm to analyse the parameters: mean nuclear size, nuclear density, mean nuclear red- green-blue (RGB) intensity and mean positive intensity. Results: No statistically significant association was found between the morphometric parameters, mean nuclear dimension, nuclear density, mean nuclear RGB intensity, and mean positive intensity, with diagno- sis, recurrence, or staging. However there seems to be an association with clinical significance, namely of the parameters mean positive intensity and mean nuclear size. In these parameters there seem to be differences between the non-invasive or intraepithelial stages (pTis and pTa), compared to cases with suburothelial connective tissue inva- sion (pT1). Mean nuclear size is higher and mean positive intensity is lower in invasive T1 cases.