Dottorato in FISIOPATOLOGIA ED IMAGING CARDIO-TORACO-VASCOLARE

Titolo della tesi: CT Radiomics-Based Machine Learning for Predicting Treatment Response in unresectable stage III-IV NSCLC: Insights from Dual Chemo-Immunotherapy and Immunotherapy Cohorts

Background: early prediction of treatment response in advanced-stage unresectable non-small cell lung cancer (NSCLC) is essential for optimizing therapeutic decisions. Radiomics combined with machine learning provides a promising non-invasive approach for quantifying tumor heterogeneity and improving early response assessment. Objective: the aim of this study is to develop a machine learning model based on radiomic features extracted from baseline contrast-enhanced CT scans for predicting early treatment responses in two cohorts of patients with stage III-IV NSCLC treated with distinct therapeutic regimens (first-line combined chemo-immunotherapy or immunotherapy alone). Methods: in this prospective, bicentric, two-cohort study, patients with confirmed diagnosis of advanced non–small cell lung cancer (stage III or IV) were prospectively enrolled. At Center 1, after applying exclusion criteria, 90 eligible patients (52 CHT/IT and 38 IT) were included for model development and internal validation. An independent external cohort from Center 2 included 40 additional patients, 20 treated with CHT/IT and 20 with IT—used for external validation. Radiomic features were extracted from segmented tumor volumes using the Trace4Research™ platform and multiple machine learning models (Random Forest, Support Vector Machine, K-Nearest Neighbors, Multi-Layer Perceptron, Logistic Regression) were trained and compared. Results: in the CHT/IT cohort (Group A), seven radiomic features were retained after outlier removal. The support vector machine (SVM) classifier achieved the highest performance (ROCAUC = 0.90 [0.82–0.97]; accuracy = 82%; sensitivity = 82%; specificity = 83%; PPV = 67%; NPV = 94%; p < 0.005). For the IT cohort (Group B), 295 features were analyzed, and the SVM again provided optimal performance (ROC-AUC = 0.84 [0.73–0.95]; accuracy = 77% ; sensitivity = 82%; specificity = 70%; PPV = 80%; NPV = 83% ; p < 0.05).External validation using 20 independent cases per group confirmed the robustness and generalizability of both models. Conclusions: Machine learning models leveraging CT-derived radiomic features enables accurate, non-invasive prediction of early response in patients with advanced NSCLC treated with either chemo-immunotherapy or immunotherapy alone. These findings support radiomics as a promising non-invasive imaging biomarker for personalized treatment planning and therapeutic outcome evaluation in lung cancer.