Endocrine Abstracts

Prostate cancer is multifocal in nature, and histological grading is the key clinical prognostic factor. To build non-subjective histological analysis tools, and to model the multifocality of prostate cancer within the organ, we use analysis of histological images to quantitatively describe prostate cancer. Our current effort shows how heterogeneity in prostate tissue due to cancer or spatial location can be quantified with image-derived features. We use high-resolution digital whole slide images of serial sections of H&E stained tissue, enabling to use data from whole organ for quantitative analysis. We use automated image analysis for extracting several (hundreds) local descriptors capturing the characteristics of each spatial location. The descriptors include several common image morphology, texture and intensity features as well as features specifically engineered for characterizing the spatial context of the region. We then use these descriptors for building a discriminative model for normal and cancerous tissue, as well as for separating spatial locations within prostate. Specifically, we use machine learning for building a classifier model, yielding a subset of informative features related to spatial heterogeneity. Our hope is that this will lead to increased knowledge of the histological changes in prostate. Our aim is to histologically model prostate cancer in 3D, and in the future, combine sequencing based measurements in the three-dimensional spatial context. Furthermore, we will correlate genomic measurements with histological features.

Presenting Author: Pekka Ruusuvuori, BioMediTech, University of Tampere, FI-33014, Tampere, Finland. Email: [email protected]