In:
Cancer Research, American Association for Cancer Research (AACR), Vol. 82, No. 12_Supplement ( 2022-06-15), p. 5061-5061
Abstract:
Most primary prostate cancers are multifocal, with multiple genomically independent tumors identified in up to 80% of men undergoing radical prostatectomy for clinically localized disease. The heterogeneity in prostate cancer and the clinical significance of secondary tumors have been explored. However, studies to date have been limited to analyzing prostate biopsies or sections using dominant tumor foci only, not allowing for the evaluation of the inter and intra tumor molecular heterogeneity of the prostate cancer landscape. Recently, we identified mutually exclusive expression patterns of more than one driver molecular aberration in distinct tumor foci. Here, we present a quantitative analysis of tumors with distinct markers and correlate with clinical outcomes. We developed a model to quantify cancerous regions and genotype expression on immunohistochemical (IHC) slides. Our algorithm combines multiple modeling strategies, including deep neural networks and color deconvolution. Dual ERG/SPINK1 IHC staining was performed on whole mount prostatectomy slides. Patches from corresponding slides stained with Hematoxylin/eosin (H & E) were generated to train the model. A color devolution method was used to separate the patches into H & E and other stain channels. For evaluation, patches were generated from IHC slides, which were then transformed into Hematoxylin-only stained patches. Sample patches were analyzed for detection of cancerous regions at the pixel-level. The ratios of cancerous regions were calculated. Our segmentation results were compared with the area positive for ERG or SPINK1. For the ERG stain, the model demonstrated a sensitivity of 62.72% per pixel, whereas it was associated with a sensitivity of 48.37% per pixel for SPINK1. This represents a robust performance for a proof-of-concept model, and sensitivities are expected to improve significantly with additional annotations. To investigate the entire landscape of whole mount samples, training to include recognition of morphological variations such as HGPIN is underway and expected to also increase sensitivity. More detailed analysis will be shared at the time of presentation. We present here a quantitative analysis model that reveals unprecedented details on tumor heterogeneity with respect to molecular markers and tumor localization. To our knowledge, this is the first study on quantitative molecular mapping of cancer and biomarker expression using whole mount samples. The marker expression in each tumor foci can be correlated with clinicopathologic findings. Different molecular subtypes presented with more than one driver molecular aberration in distinct tumor foci may be associated with distinct clinical outcomes such as biochemical recurrence or metastasis, allowing for predicting disease progression and targeted approaches for effective prostate cancer management. Citation Format: Joonyoung Cho, In Hye Suh, Tae-Yeong Kwak, Sun Woo Kim, Hyeyoon Chang, Nallasivam Palanisamy. Molecular mapping of prostate cancer on whole mount prostatectomy specimens using deep neural networks to quantify genotypic heterogeneity [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5061.
Type of Medium:
Online Resource
ISSN:
1538-7445
DOI:
10.1158/1538-7445.AM2022-5061
Language:
English
Publisher:
American Association for Cancer Research (AACR)
Publication Date:
2022
detail.hit.zdb_id:
2036785-5
detail.hit.zdb_id:
1432-1
detail.hit.zdb_id:
410466-3
