Abstract B077: 3D reconstruction and machine learning-based analysis of prostate cancer from histologic images

Abstract

Abstract Prostate cancer is multifocal in nature, and histologic grading is the key clinical prognostic factor. Imaging-based tools are required for decreasing the subjectivity of histologic grading, providing quantitative information of the tissue and the pathologic changes in it, and to allow quantitative associations of histologic alterations with other types of information collected from the tissues, such as genomic data. To enable imaging-based diagnosis, methods for quantifying both the naturally occurring heterogeneity of normal tissue and morphologic changes due to pathology need to be developed. On the other hand, accurate distinction of early pathologic changes from natural variation could provide novel information about development of tumors. Furthermore, the 3-dimensional evolution and growth patterns of tumors should be considered as, traditionally, histologic scoring mostly relies on individual tissue sections extracted from their spatial context. To build nonsubjective histologic analysis tools, and to model the multifocality of prostate cancer within the organ, we use analysis of histologic images to quantitatively describe prostate cancer. We present an approach for 1) imaging the whole organ into digital pathology slides, 2) reconstruction of the 3D structure of the organ based on the histologic images, and 3) both feature-based and deep learning-based quantitative analysis of the digital images. Our approach enables characterization of tissue morphology with numerical descriptors, enabling subsequent analysis, such as determining the likelihood of pathologic changes. Our current efforts show how heterogeneity in prostate tissue due to spatial location or cancer can be quantified with image-derived features. In addition, we use mouse prostate as a model to visualize and reconstruct a whole organ from high-resolution whole-slide images, and combine tissue type classification in the 3D reconstruction. Further development of these methods and their application for human samples will improve our understanding of human prostate pathologies and cancer evolution within the 3D environment of the organ. Citation Format: Pekka Ruusuvuori, Mira Valkonen, Kimmo Kartasalo, Tapio Visakorpi, Matti Nykter, Leena Latonen. 3D reconstruction and machine learning-based analysis of prostate cancer from histologic images [abstract]. In: Proceedings of the AACR Special Conference: Prostate Cancer: Advances in Basic, Translational, and Clinical Research; 2017 Dec 2-5; Orlando, Florida. Philadelphia (PA): AACR; Cancer Res 2018;78(16 Suppl):Abstract nr B077.