ECP 2022 Abstract Book

Virchows Archiv (2022) 481 (Suppl 1):S1–S364 13 PS-07-038 Agreement in tumour budding and poorly differentiated cluster detection in consecutive digitalized images with focus on mim- ickers, serial images and expertise M. Jimeno*, R. López-Martos, J.P. Fernández, M.Á. Serrano, S. Roca, A. Sánchez-Gil *Hospital Universitari Germans Trias i Pujol, Spain Background & objectives: Tumour budding (TB) is a recognized adverse prognostic factor in colorectal cancer (CRC) but agree- ment on its quantification varies significantly. We aim to study interobserver agreement using digitalized slides with an emphasis in mimickers, serial images and observers’ experience. Methods: Three consecutive H&E - CK AE1.3 - H&E digitalized images were obtained from 30 CRC cases, using a “hotspot” study area of 0.785 mm2 for each. Three observers with differing experience reviewed the 90 images, reporting the number of TB and poorly differentiated clusters (PDF) in addition to false positive images. For agreement analysis, the intraclass correlation coefficient was applied. Results: Analysis showed moderate agreement values for TB and PDC detection in H&E images. Despite the presence of abundant mimickers in CK sections, TB and PDC detection was facilitated, and a higher correlation was observed, especially for cases with moderate or severe inflammation. No significant disparity in slide assessment was evident between the 3 observers. Consecutive sections have shown a significant evolution of images with variable size and number of small groups of cells in only 10 microns thick. A notable number of cases have shown a significant progression towards a lower or higher TB/PDC number, that is enough to modify the final grade. Conclusion: Despite reports of moderate to substantial agreement with respect to TB or PDC grade, agreement with respect to individual cell groups is moderate at best. Most studies published to date have selected representative slides but no specific area to be evaluated, leading to intrinsic selection bias. Cell morphology evolution observed in serial images evidence the need of algorithm-based analyses of larger areas to obtain an average value that could be translated into a powerful and reproducible prognostic tool. PS-07-039 Automated tumour budding quantification in T1 colorectal carcinoma H&E slides: association to lymph node metastasis I. Nearchou*, Y. Kajiwara, M. Ueno, K. Kouzu, T. Nakamura, K. Lillard, H. Ueno *Indica Labs, Inc., USA Background & objectives: Tumour budding (BD) is a significant predictor of lymph node metastasis (LNM) in T1 colorectal cancer (CRC). However, the reproducibility of BD-scoring remains sub- optimal. This study aimed to automatically quantify BD on H&E slides and assess its association with LNM. Methods: A deep-learning algorithm was applied on 197 T1 CRC H&E slides to automatically detect BD using HALO® and HALO AI™ image analysis platform. Univariate and multivariate logistic regressions were performed to assess the predictive value of BD and other features currently assessed in the clinic. The Akaike information criterion was used to identify the model with the high- est predictive value. Results: Various automated BD quantification methods were employed including recording the number and density of buds across the entire invasive front (IF) and at a hotspot area. BD num- ber and density assessed on all slides containing the IF (p=0.001 and p=0.001) as well as on a single slide with highest BD (p=0.004 and p=0.013) were shown to be significantly associated with LNM when assessed using univariate logistic regression. The model with highest predictive value for LNM included BD density assessed on all slides, which was also the most significant feature (p<0.001), tumour grade (p<0.001) , lymphovascular invasion (p=0.285) and submucosal involvement depth (p=0.044). Conclusion: Here, we demonstrate that the use of deep-learning algorithms could prove to be promising for the objective, standard- ised and reproducible BD quantification in H&E slides as well as for assisting pathologists during treatment decision making. Funding: This study was funded by the Japan Society for the Pro- motion of Science. Indica Labs, Inc. provided in-kind resources. PS-07-040 Inter-observer concordance in the measurement of histological risk factors in pT1 colorectal adenocarcinomas K. Saez De Gordoa*, I. Archilla, M. Rodrigo-Calvo, J.J. Aguirre, M. Camara Jurado, A. Canosa, F. Giner, M. Jimeno, I. Jurado, S. Landolfi, A. Lloret, I. Machado, C. Martínez-Ciarpaglini, E. Musulen, D. Naranjo, N. Papaleo, Ò. Rosiñol, R. Sanchez Yuste, G.T. Vázquez Benítez, M. Cuatrecasas *Pathology Department, Hospital Clínic, Barcelona, Spain Background & objectives: Identification of risk factors associated with lymph node metastasis (LNM) in pT1 colorectal carcinoma (CRC) is performed by histological evaluation. The aim of this study is to assess the inter-observer agreement in the evaluation of histological risk factors for LNM. Methods: Scanned slides from 10 pT1 CRCs were assessed by 22 pathologists for submucosal infiltration depth (SID), lymphovas- cular invasion (LVI), histological grade (HG), and tumour budding (TB). The LVI, HG and TB concordance were calculated with percentage of agreement and Krippendorff’s alpha; intraclass cor- relation coefficient (ICC) for SID. An ulterior consensus meeting was held to analyse differences in histologic criteria assessment. Results: Seven tumours were resected by endoscopy, and three by surgery. Four pT1 arose on pedunculated polyps, and six on sessile polyps. The evaluation of SID had moderate reliability, with an average ICC (two-way random-effects model) of 0.59 (95% confidence interval 0.12-0.88). LVI had the highest agreement rate (83.64%). TB had 71.36% of agreement and Krippendorff’s alpha (α) of 0.13. HG had 64.44% of agreement and α = 0.26. In the consensus meeting the use of the measuring tool was corrected, as it generated discordances. An agreement on the criteria for assessing SID for pedunculated and sessile polyps was also reached, and the criteria for LVI, TB and HG were reviewed. Conclusion: This is a pilot ongoing study. Lack of consistency in the evaluation of some prognostic factors by pathologists was mostly due to differences in measurement criteria, as well as the small number of cases evaluated by many pathologists. PS-07-041 Diagnostic utility of CK20, SATB2, CDH17, and Villin for the identification of gastrointestinal tumour origin: a tissue microarray study on 7,711 tumours of 117 tumour entities D. Dum, P. Lebok, M. Lennartz, S. Weidemann, E. Burandt, C. Fraune, R. Simon, T. Krech, F. Büscheck, T.S. Clauditz, F. Jacobsen* *University Medical Center Hamburg, Germany Background & objectives: Adenocarcinomas of the gastrointestinal tract (GIT) represent a common source of liver and lung metastasis. Markers used to distinguish metastases from GIT tumours include CK20, SATB2, CDH17, and villin. S101