Subjective analyses of images by pathologist reviewers are plagued by issues of inter-rater variability and throughput. However, as digital pathology whole slide scanners become more commonplace, the amount of high-quality pathology image data available to researchers and clinicians is increasing, and the newfound widespread availability of pathology images in digital form, including the NCI Cancer Genome Atlas (TGCA), opens up new possibilities to use computational approaches to leverage the information inherent within them for diagnosis, prognosis, and precision medicine. This award supports initiation of a collaborative research project that aims to discover new quantitative image-based prognostic biomarkers for prostate cancer, focusing on an investigation of novel concepts from computational topology applied to prostate cancer glandular architecture.
The current standard for prostate cancer grading is the Gleason score, which is a subjective rating system based on an analysis of high-level tissue architecture and glandular shape and organization. However, Gleason scoring is variable between pathology reviewers, and may not capture all of the potentially prognostic information contained in glandular growth patterns. In this project, new topological descriptors will be developed that capture architectural features of prostate glands in pathology images. These descriptors can then be used to aid pathologists by providing more quantitative and more reproducible analogs to the traditional Gleason scores, and they may have independent prognostic value. They can also be used to classify slides in order to distinguish between different types of cancerous architectures of glands, compared to the current gold-standard histopathological and molecular characterization. In particular, the aim of this project is to demonstrate effectiveness of using computational methods based on tools from computational geometry and topology to recognize and quantify glandular architectural features. Glandular density will be the first architectural feature quantified in this collaborative work. This award is supported by the National Institutes of Health Big Data to Knowledge (BD2K) Initiative in partnership with the National Science Foundation Division of Mathematical Sciences.
