Official websites use. Share sensitive information only on official, secure websites. This study developed and validated a novel deep learning radiomic biomarker to estimate response to immune checkpoint inhibitor ICI therapy in advanced nonโsmall cell lung cancer NSCLC using real-world data RWD and clinical trial data.
Biomarker generalizability was further evaluated using a secondary analysis of a prospective clinical trial ClinicalTrials. The primary aim of this study was to develop an imaging-based biomarker to identify patients who will benefit from PD- L 1 immune checkpoint inhibitor ICI as a monotherapy in advanced nonโsmall cell lung cancer NSCLC.
Can a biomarker on the basis of routinely acquired pretreatment computed tomography CT scans provide prognostic value, complimentary to other predictors, and generalize to out-of-cohort validation data? Our study developed and validated a deep learning biomarker using a multi-institute real-world data set. The biomarker identified patients with progression-free survival and overall survival OS benefits from PD- L 1 ICI in a real-world cohort and OS benefit in a clinical trial validation cohort, demonstrating the potential for a clinical tool that can enhance treatment decision making.
Three biopsy-based biomarkers are US Food and Drug Administrationโapproved for ICI treatment guidance: PD-L1 immunohistochemistry, 7 tumor mutation burden, 8 and microsatellite instability high status. These biomarkers have demonstrated limited predictive power, 11 which may stem from inter- and intratumor heterogeneities and the inherently limited predicted power of a single biomarker. Biomarkers on the basis of medical imaging features ie, radiomics have been investigated for ICI response prediction.
Radiomic biomarkers can also evaluate the overall tumor burden and may identify predictive signal in regions outside of the tumor. Traditional radiomics with hand-crafted features typically involve segmentation of regions of interest and extraction of hundreds to thousands of predefined features that quantify various components of shape, intensity, and texture, which are subsequently processed using machine learning algorithms.
