Towards Integrative Data Analysis for Predictive Modeling in Biomedical Computing

这项研究旨在开发创新方法来分析医疗数据，并了解复杂疾病，特别是前列腺癌（PCa）的作用机制。大规模生物学的新时代的特征是不可克服的数据量，包括医学图像，高通量遗传信息，大型公共数据库和临床信号。数据丰富与我们对疾病背后的生物事件的有限知识形成鲜明对比。拟议的研究旨在通过一种新的框架来缩小这一差距，该框架使用最先进的图像分析方法，并整合分子数据和生物信息，以制定更准确的模型来表征癌症。 虽然PCa是北美男性中第三大癌症相关死亡原因，但如果早期诊断，它可以以90%的长期无病生存率进行管理。PCa的诊断需要在超声引导下进行活检。然而，超声引导不是针对个人的，并且在检测癌症，特别是其侵袭性形式方面的灵敏度很差。迫切需要明确区分需要干预的侵袭性PCa和可以被置于“警惕等待”的惰性PCa。本文开发的算法将通过以下方式解决这一挑战：i）使用来自概率图的信息增强超声图像，这些信息指示癌症的侵袭性; ii）对分子信息进行综合分析，以分层疾病结果;以及iii）整合多分辨率数据的特征，以构建PCa的综合模型。 超声信号的采集和分析的无与伦比的方法提供了增强的预测PCa的存在和程度。通过整合多种模式的特征而制定的创新模型促进了个性化治疗和活检的发现。建立的方法和软件将帮助临床医生做出更好的管理决策，以降低其死亡率，同时平衡不必要的过度治疗的风险。所寻求的计算方法将推动图像和信号处理、机器学习、系统生物学和软件工程的研究。在PCA的框架中讨论了方法;然而，它们可以扩展到其他疾病。 高素质人才的培训是这项工作的主要重点;三名博士生和三名硕士生，以及五名暑期本科生将参与这项研究计划。

This research is aimed at developing innovative methods to analyze medical data and understand the mechanisms of actions of complex disease, specifically prostate cancer (PCa). The new era of large-scale biology is characterized by insurmountable amounts of data including medical images, high throughput genetic information, large public databases, and clinical signals. Data abundance stands in contrast to our limited knowledge about the biological events underlying disease. The proposed research seeks to close this gap through a novel framework that uses state-of-the-art image analysis methods, and integrates molecular data and biological information, to formulate more accurate models to characterize cancer. Although PCa is the third leading cancer-related cause of death in North American men, if diagnosed early, it can be managed with a 90% long-term disease free survival rate. Diagnosis of PCa requires biopsy, guided by ultrasound. However, ultrasound guidance is not tailored to an individual and has poor sensitivity in detection of cancer, specifically its aggressive forms. There is an urgent need to clearly distinguish between aggressive PCa that requires intervention and indolent PCa that can be put on “watchful waiting”. The algorithms developed here will address this challenge by i) augmenting ultrasound images with information from probability maps indicating the aggressiveness of cancer, ii) comprehensive analysis of molecular information to stratify disease outcome; and iii) integration of features at multiple resolutions of data to build a comprehensive model of PCa. The unparalleled methods for acquisition and analysis of ultrasound signals provide enhanced prediction of the existence and extent of PCa. The innovative models formulated by integration of features of multiple modalities facilitate the discovery of personalized therapies and biopsies. The methods and software built will help the clinicians make better management decisions for PCa to reduce its mortality while balancing it against the risks of unnecessary overtreatment. The computational methods sought will advance research in image and signal processing, machine learning, systems biology and software engineering. Methods are discussed in the framework of Pca; however, they are extendible to other disorders. Training of highly qualified personnel is a primary focus of this work; three PhD and three Master’s students, as well as five summer undergraduates will be involved in this research program.