Computed Tomography-Based Delta-Radiomics Analysis for Discriminating Radiation Pneumonitis in Patients With Esophageal Cancer After Radiation Therapy

Purpose Our purpose was to construct a computed tomography (CT)–based delta-radiomics nomogram and corresponding risk classification system for individualized and accurate estimation of severe acute radiation pneumonitis (SARP) in patients with esophageal cancer (EC) after radiation therapy. Methods and Materials Four hundred patients with EC were enrolled from 2 independent institutions and were divided into the training (n = 200) and validation (n = 200) cohorts. Eight hundred fifty radiomics features of lung were extracted from treatment planning images, including the positioning CT before radiation therapy (CT1) and the resetting CT after receiving 40 to 45 Gy (CT2). The longitudinal net changes in radiomics features from CT1 to CT2 were calculated and defined as delta-radiomics features. Least absolute shrinkage and selection operator algorithm was performed to features selection and delta-radiomics signature building. Integrating the signature with multidimensional clinicopathologic, dosimetric, and hematological predictors of SARP, a novel CT-based delta-radiomics nomogram was established according to multivariate analysis. The clinical application values of nomogram were both evaluated in the training and validation cohorts by concordance index, calibration curves, and decision curve analysis. Recursive partitioning analysis was used to generate a risk classification system. Results The delta-radiomics signature consisting of 24 features was significantly associated with SARP status (P < .001). Incorporating it with other high-risk factors, Subjective Global Assessment score, pulmonary fibrosis score, mean lung dose, and systemic immune inflammation index, the developed delta-radiomics nomogram showed increased improvement in SARP discrimination accuracy with concordance index of 0.975 and 0.921 in the training and validation cohorts, respectively. Calibration curves and decision curve analysis confirmed the satisfactory clinical feasibility and utility of nomogram. The risk classification system displayed excellent performance on identifying SARP occurrence (P < .001). Conclusions The delta-radiomics nomogram and risk classification system as low-cost and noninvasive means exhibited superior predictive accuracy and provided individualized probability of SARP stratification for patients with EC. Our purpose was to construct a computed tomography (CT)–based delta-radiomics nomogram and corresponding risk classification system for individualized and accurate estimation of severe acute radiation pneumonitis (SARP) in patients with esophageal cancer (EC) after radiation therapy. Four hundred patients with EC were enrolled from 2 independent institutions and were divided into the training (n = 200) and validation (n = 200) cohorts. Eight hundred fifty radiomics features of lung were extracted from treatment planning images, including the positioning CT before radiation therapy (CT1) and the resetting CT after receiving 40 to 45 Gy (CT2). The longitudinal net changes in radiomics features from CT1 to CT2 were calculated and defined as delta-radiomics features. Least absolute shrinkage and selection operator algorithm was performed to features selection and delta-radiomics signature building. Integrating the signature with multidimensional clinicopathologic, dosimetric, and hematological predictors of SARP, a novel CT-based delta-radiomics nomogram was established according to multivariate analysis. The clinical application values of nomogram were both evaluated in the training and validation cohorts by concordance index, calibration curves, and decision curve analysis. Recursive partitioning analysis was used to generate a risk classification system. The delta-radiomics signature consisting of 24 features was significantly associated with SARP status (P < .001). Incorporating it with other high-risk factors, Subjective Global Assessment score, pulmonary fibrosis score, mean lung dose, and systemic immune inflammation index, the developed delta-radiomics nomogram showed increased improvement in SARP discrimination accuracy with concordance index of 0.975 and 0.921 in the training and validation cohorts, respectively. Calibration curves and decision curve analysis confirmed the satisfactory clinical feasibility and utility of nomogram. The risk classification system displayed excellent performance on identifying SARP occurrence (P < .001). The delta-radiomics nomogram and risk classification system as low-cost and noninvasive means exhibited superior predictive accuracy and provided individualized probability of SARP stratification for patients with EC.