Delta total tumor volume and radiomics as predictive biomarkers in microsatellite instability-high metastatic colorectal cancer

ObjectivesMicrosatellite instability-high (MSI-H) metastatic colorectal cancer (mCRC) patients show heterogeneous responses to immunotherapy, and early predictive biomarkers are lacking.This study aims to evaluate whether variations (delta-) in total tumor volume (TTV) and radiomic features can early predict Overall Survival (OS) in patients with mCRC with MSI-H undergoing immunotherapy. Material and MethodsOne hundred forty-five patients diagnosed with MSI-H mCRC (mean age, 56 ± 15 years; 89 men) were retrospectively included from two centers.Radiologists manually annotated all tumors (primary and metastatic lesions) on chest-abdomen-pelvis computed-tomography at baseline (BL) and first evaluation (E1) scans.The median time between BL and E1 was 50 days (interquartile range: 45 -64 days).The median survival time was not reached, the median follow-up time was 61.34 months.In total, 73.79% (107/145) patients were censored.The first censor happened at 10.8 months (interquartile range: 38.4 -75.1 months).Delta-radiomic features were combined with delta-TTV using machine learning models to predict OS.The ability to stratify patients into long-and short-term survival groups was assessed using the Restricted Mean Survival Time (RMST) ratio, area under the curve (AUC) at 3 years, and concordance index (C-index).Generalization was assessed through 75 train/test cohort splits. ResultsA total of 2 079 lesions at BL and 2 112 at E1 were annotated.Delta-TTV outperformed TTV at BL and E1 across all metrics (p17.3% TTV increase had significantly worse survival.Combining delta-radiomics with delta-TTV enhanced performance, improving the RMST ratio from 2.215 to 2.450 (2.248-2.652,p=0.019), the AUC from 0.854 to 0.875 (0.862-0.888, p