ECP 2023 Abstracts
S233 Virchows Archiv (2023) 483 (Suppl 1):S1–S391 13 Background & objectives: The mean of ten largest nucleoli in uveal melanoma is an important prognosis factor, however calculating this parameter is challenging and imprecise with usual methods. We pro- pose an artificial intelligence automated method to evaluate correctly and easily this prognosis factor. Methods: We chose a batch of 1280 photos, we did a training of the data then de-normalization “image processing” to finish with a display of the batch. then we had recourse to the discriminator of a generative adversarial networks which was simply a classifier. Li distinguish the real data from those created by the generator. Results: For image processing, we applied the, cv2.COLOR_BGR- 2RGB, we have eliminated the colours of the image to leave only the levels of grey. Then we set the limits of the image, we searched for the colours within the specified limits and applied the mask then we applied the Blur to blur the image. Then we made the outline of the nucleoli. We counted the number of closed contours, which is the num- ber of total cells. this allowed us to be able to calculate the diameter of the nucleoli and to keep only the 10 largest cells and average the sum of their diameter. Conclusion: Automated methods using artificial intelligence are an effective and time saving tool, that can be used in histopathology for correct and precise evaluation of prognosis factors. E-PS-08-006 Smart system for determining histological differentiation of pros- tate cancer biopsy using artificial intelligence techniques: from unsupervised to supervised learning S. Bhattacharjee*, Y. Hwang, N. Cho, D. Kim, H. Kim, H. Choi *Inje University, Republic of Korea Background & objectives: Analysis of histological differentiation of prostate cancer biopsy can be challenging but early detection can help with treatment planning by knowing the cancer stage. We aimed to develop a smart system for diagnosing prostate cancer using artificial intelligence techniques. Methods: Whole slide image (WSI) datasets from Yonsei University Hospital (YUH), Korea, and Radboud University Medical Center (RUMC), Netherlands, were used in this study. We performed unsu- pervised to supervised learning using the modified K-means clustering algorithm and newly developed deep learning (DL) model, respectively, on the internal dataset (YUH). Furthermore, the trained DL model was tested on the external dataset (RUMC). Results: The unannotated WSIs were used for unsupervised learning to generate the labelled patch images for supervised learning. To train the model efficiently, we used the pre-trained weight of self-supervised learning. However, the DL model showed almost perfect agreement for internal data (quadratic weighted Kappa 0.872; 95%CI 0.815-0.928) and external data (quadratic weighted Kappa 0.835; 95%CI 0.766- 0.904) in predicting stroma, benign, and cancer tissue components. To analyse the generalizability of our model and evaluate qualitative outputs, we compared the results with the annotated samples from YUH and RUMC. The model showed promising performance though few mispredictions were observed, this can be justified due to the use of an unsupervised technique. Conclusion: In this study, our approach provides annotation-free prostate cancer biopsy diagnosis distinguishing between stroma, benign, and cancer. Moreover, the AI-based smart system has been applied in multiclass classification for determining histological dif- ferentiation of prostate cancer biopsy. The combination of unsuper- vised, self-supervised, and supervised learning can be a promising approach that could mitigate the problem of label-free data classi- fication. In the future, we will extend this research to multi-cancer datasets and explore our approach to overcome the challenges of unsupervised learning. Funding: This research was supported by the National Research Foun- dation of Korea (NRF) grant funded by the Korea government (MIST) (Grant No. 2021R1A2C2008576), and a grant from Korea Health Tech- nology R&D Project through the Korea Health Industry Development Institute validation (KHIDI), funded by the Ministry of Health & Wel- fare, Republic of Korea (Grant No: HI21C0977). E-PS-08-007 Upregulation of TIM3 and reduced expression of PD-1 on immune cell subsets in advanced prostate cancers N. Blessin*, N.F. Debatin, E. Bady, J.H. Müller, T. Mandelkow, R. Simon, M. Lennartz, C. Bernreuther, Z. Huang, G. Sauter, T.S. Clau- ditz, S. Minner, E. Burandt, M. Graefen, N. Gorbokon *Institute of Pathology, University Medical Center Hamburg-Eppen- dorf, Germany Background & objectives: Although most prostate cancers behave in an indolent manner, a small proportion is highly aggressive. Both pri- mary and advanced prostate cancer is widely known as a non-inflamed cancer that is characterized by a paucity of immune infiltration. Methods: To assess the spatial interplay of more than 30 TIM3, CTLA-4, PD-1/-L1 expressing leukocyte subpopulations in 453 pros- tate cancers, tissue microarrays were stained with 21 antibodies using our BLEACH&STAIN multiplex fluorescence immunohistochemistry approach and analysed using a deep learning-based image analysis framework. Results: The immune cell density of CD8+ cytotoxic T-cells, CD4+ T-helper cells, FOXP3+ regulatory T-cells, M1/ M2 macrophages, as well as CD11c+ dendritic cells increased consistently along with the Gleason grade in primary prostate cancer (p≤ 0.034 each). In recur- rent prostate cancers under therapy, the density of FOXP3+ regulatory T-cells and M1 macrophages further increased, while the density of CD8+ cytotoxic T-cells, CD4+ T-helper cells, as well as CD11c+ den- dritic cells decreased (p≤0.017 each). Although the immune checkpoint expression of TIM3 on T-cell subsets, macrophages and dendritic cells was upregulated in advanced/ recurrent tumours, the expression level of PD-1 was downregulated in all analysed T-cell subsets. Conclusion: Although prostate cancer is a generally considered a low inflamed tumour, the degree of tumour infiltration by various immune cell subtypes increases markedly along with tumour progression and in recurrent tumours under therapy. Taken together, these data suggest that the evaluation of the spatial distribution of immune cell types along with their immune checkpoint expression can provide relevant clinical information in prostate cancer. E-PS-08-008 HPyloriDet: a clinically deployable tool for computer-aided heli- cobacter pylori detection in immunohistochemically stained slides N. Brandt*, A. Bornand, D.G. Puppa, M. Kreutzfeldt, D. Merkler, A. Janowczyk *Department of Pathology, Division of Clinical Pathology Geneva University and University Hospitals, Switzerland Background & objectives: Helicobacter Pylori (HP) is a common stomach bacteria linked to conditions including stomach cancer. Although immunohistochemical (IHC) staining of HP facilitates diag- nosis, reviewing whole slide images (WSI) remains time-consuming. We investigated our computer-aided screening tool, HPyloriDet, for improving diagnostic speed/sensitivity. Methods: HPyloriDet was developed using pathologist-annotated HP IHC WSIs (n=20). Regions of interest were identified via IHC stain deconvolution, driving the extraction of 300x300 pixel patches split into 80/20 train/test split. A DenseNet was trained to detect the pres- ence of HP on these patches, and its performance metrics evaluated.
Made with FlippingBook
RkJQdWJsaXNoZXIy Mzg2Mjgy