Novel Radiomics for Predicting Response to Immunotherapy for Lung Cancer

点击翻译按钮获取中文摘要

ABSTRACT: In 2019, an estimated 228,150 patients in the US are expected to be diagnosed with non-small cell lung cancer (NSCLC). A recent landmark development has been the approval of the immune checkpoint inhibitors (anti-PD-1 and anti-PD-L1) for the treatment of locally advanced and metastatic NSCLC. These immunotherapy (IO) drugs have an excellent toxicity profile and have the potential to induce durable clinically meaningful responses. However, only 1 in 5 NSCLC patients treated with IO will have a favorable response. Unfortunately, the current tissue based biomarker approach to selecting patients for these treatments is sub- optimal due to the dynamic nature of the interaction of the immune system with the tumor. Given the prohibitive costs associated with IO (>$200K/year per patient), there is a critical unmet need for predictive biomarkers to identify which patients will not benefit from IO. Additionally, the current clinical standard to evaluating tumor response (i.e. RECIST and irRC which evaluate change in tumor size and nodule disappearance) is sub-optimal in evaluating early clinical benefit from IO drugs. This is due at least in part to the fact that some patients undergoing IO present apparent disease progression (pseudo-progression) on post-treatment CT scans. Unlike the standard canon of radiomics (computer extracted features from radiographic scans) that assess textural or shape patterns, our group has been developing novel computer vision strategies to capture patterns of peri-tumoral heterogeneity (outside the tumor) and tumor vasculature from CT scans. In N>300 patients, our group has shown that (1) radiomics of vessel tortuosity on baseline, pre-treatment CT for NSCLC patients undergoing IO were significantly different between responders (less tortuous) and non-responders (more tortuous), (2) serial changes in these measurements were better predictors of early response to IO compared to clinical response criteria such as RECIST and irRC and (3) these radiomic attributes were associated with PD-L1 expression and degree of tumor infiltrating lymphocytes on baseline biopsies. Critically, these radiomic features predicted response for NSCLC patients treated with 3 different IO agents from 3 sites. In this project we will further develop vasculature, peri- and intra-tumoral radiomic features for monitoring and predicting benefit and early response for NSCLC patients treated with IO. We will uniquely train our radiomics using a set of N>180 resected NSCLC patients treated with first line IO and for whom we will have major pathologic response (MPR) as primary endpoint. In addition, we will establish the biological underpinnings of these predictive radiomic signatures by evaluating their association with the morphology, immune landscape (from biopsies) and molecular pathways of the tumor. In addition we have access to N>700 NSCLC patients treated on completed clinical trials via our industry partners (Astrazeneca, Bristol-Myers Squibb) for tool validation. Finally, we will deploy LunIOTx within the ECOG-5163 (INSIGNA) trial (N>600), the first time that radiomics will be evaluated within a prospective cooperative group clinical trial for IO.