Rad-path-omic tools for rectal cancer treatment evaluation

点击翻译按钮获取中文摘要

PROJECT SUMMARY: Of the estimated 43,030 patients who will be newly diagnosed with rectal cancer in 2018, a majority will receive neoadjuvant chemoradiation (NAC) to reduce tumor burden. All patients ultimately undergo an aggressive excision of the rectum, of which 25% exhibit complete pathologic response (pCR, i.e. disease- free after NAC) on the post-surgical specimen. These patients have therefore been subjected to an unnecessary, morbid procedure resulting in quality of life issues, in the absence of any definitive, non-invasive biomarkers for NAC response in vivo. While multi-parametric MRI is utilized to pre-operatively assess tumor response and regression to NAC, expert interpretation is confounded and variable due to overlapping appearance of benign treatment effects (e.g. fibrosis, ulceration) and residual tumor. Recently, more quantitative characterization of lesions has been enabled via radiomics, involving high- throughput, computerized extraction of textural or kinetic attributes from imaging. Radiomic maps of the tumor environment can depict presence of different tissue types based on their structural and functional characteristics, visualized as regions of “low” and ‘high” feature expression. In fact, the post-NAC tumor environment on the excised rectal tissue specimen has been shown to reflect a variety and organization in different pathologic tissue types, also linked to patient prognosis and outcome. However, existing radiomic approaches only attempt to characterize the overall heterogeneity in a tissue region, as they lack “ground truth” to quantify tissue types and their organization on post-NAC MRI. A more comprehensive and accurate predictor for pCR based off multi- parametric MRI could thus be constructed by (a) quantifying the density and arrangement of structural and functional attributes on post-NAC rectal MRIs, and (b) optimizing radiomic descriptors against pathologically validated information of post-NAC tissue types on MRI, via spatial correlation with pathology. In this proposal, I will develop novel radiomic tools in conjunction with spatially co-localized “ground truth” pathology to build a predictor for identifying rectal cancer patients exhibiting pCR via post-NAC MRI. Aim 1 will involve developing and evaluating a novel radiomic descriptor to quantify spatial organization of morphologic (via structural MRI) and physiologic (via contrast enhancement functional MRI) heterogeneity of the post-NAC lesion environment. Aim 2 will focus on optimizing this radiomic organization descriptor to capture distinctive tissue organization associated with pCR, via spatial mapping of post-surgical pathology information onto pre-operative MRI. My novel descriptor will be evaluated and validated via a machine learning predictor to identify patients exhibiting pCR using a discovery and a hold-out validation cohort, acquired from 2 different institutions; and compared with clinical markers of response. My project will build on promising preliminary results for my radiomic organization descriptor as well as a radiology-pathology co-registration framework, to result in a clinically reliable and impactful radiomics-based tool which could enable personalized management of rectal cancer patients.