QuBBD: Collaborative Research: Quantifying Morphologic Phenotypes in Prostate Cancer - Developing Topological Descriptors for Machine Learning Algorithms

该项目的长期目标是开发定量方法，用于检测从（组织学）图像中提取的点云的几何和拓扑特征。特别相关的是，本项目考虑了前列腺癌分类的设置，这是基于病理学家使用Gleason分级系统对组织学切片进行分级。这些病理切片是生物医学大数据的一个来源，越来越多的人可以作为存档材料获得这些数据。发展这些定量方法将是迈向前列腺癌侵袭性（半）自动化定量的重大进展。该奖项支持计算数学、计算机科学、生物医学工程和病理学领域的跨学科研究团队开发基于拓扑描述符和机器学习的数学和计算工具，以区分前列腺癌的不同形态类型。本研究将开发定量的拓扑描述符（例如，持久性图和摘要）来描述前列腺癌的自然组织学表型，以提供解释性信息，以帮助提供改进的诊断/预后和洞察患者的最佳治疗过程。这将通过通过无监督机器学习开发图形模型来实现，从而增加我们对前列腺癌亚型的理解。长期目标是开发成像生物标志物，与现有的、主观的、可变的人类观察者分析（即Gleason评分）相比，更好地识别侵袭性前列腺癌的惰性。该项目采取步骤，为前列腺癌分类的一种新的定量方法，以及对发展拓扑方法统计区分不同类型的腺体结构。

The long-term goal of this project is to develop quantitative methodology for detecting geometric and topological features in point clouds extracted from (histology) images. Of particular relevance, this project considers the setting of prostate cancer classification, which is based on a pathologist grading of histology slides using the Gleason grading system. These pathology slides are a source of biomedical big data that are increasingly available as archived material. Developing these quantitative methods will be a significant advance towards a (semi-)automated quantification of prostate cancer aggressiveness. This award supports an interdisciplinary team of investigators in computational mathematics, computer science, biomedical engineering, and pathology to develop mathematical and computational tools based on topological descriptors and machine learning in order to distinguish between different morphological types of prostate cancer.This research will develop quantitative topological descriptors (e.g., persistence diagrams and summaries) that describe natural histologic phenotypes in prostate cancer, in order to provide explanatory information to assist in providing improved diagnostics/prognostics and insight into the best course of treatment for the patient. This will be accomplished through developing graphical models via unsupervised machine learning that increase our understanding of prostate cancer subtypes. The long-term goal is to develop imaging biomarkers that better identify indolent from aggressive prostate cancer compared to existing, subjective, and variable human observer analyses (i.e., the Gleason score). This project takes steps towards a novel quantitative methodology for prostate cancer classification, as well as towards developing topological methods for statistically distinguishing different types of glandular architectures.