Personalized Risk Stratification in Colon Cancer using Radiomic-Based Predictive Models

Authors

Abstract

Colon Cancer (CC) is among the most frequently diagnosed malignancies and a leading cause of cancer-related death worldwide. Five-year survival varies markedly by stage at diagnosis, underscoring the need for precise risk stratification. Otherwise, although stage III CC patients routinely receive adjuvant chemotherapy, some exhibit a low risk of recurrence and could safely undergo shorter regimens. Conversely, a subset of stage II patients faces a higher relapse risk and may benefit from intensified treatment. In this context, radiomics has emerged as a cutting-edge, non-invasive approach capable of extracting quantitative information from routine medical imaging to support clinical decision-making. We aimed to develop radiomic-based machine learning models able to distinguish stage II from stage III CC patients while identifying individuals at increased risk of 5-year relapse. A cohort of 104 patients who underwent preoperative computed tomography was analyzed by 3D tumor segmentation. Thus, 105 radiomic features and preoperative clinical variables were extracted. Predictive models were trained and validated employing a 70/30 split and 10-fold cross-validation. The Generalized Linear Model achieved the best performance for stage differentiation (AUC=0.760). For 5-year relapse prediction, the Partial Least Squares model showed excellent performance (AUC=0.910), outperforming the Support Vector Machine model (AUC=0.730). Subgroup analyses confirmed strong predictive capacity when evaluating stage II (AUC=0.929) and stage III patients (AUC=0.926) separately. To our knowledge, this is among the first studies demonstrating that radiomics can simultaneously stratify CC stage and predict relapse risk with high accuracy, highlighting its potential as a powerful tool to guide personalized treatment strategies. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=166 SRC="FIGDIR/small/25341095v1_ufig1.gif" ALT="Figure 1"> View larger version (28K): [email protected]@176a8c8org.highwire.dtl.DTLVardef@14322d1org.highwire.dtl.DTLVardef@a30589_HPS_FORMAT_FIGEXP M_FIG C_FIG

Tags

Topics