ESTRO 2024 - Abstract Book

S1701

Clinical - Lung

ESTRO 2024

consent for the anonymized use of data for research purposes. The segmentations of the GTV of the 4D-CT reconstructed phases, 0% and 50%, were performed in RayStation software (RaySearch Laboratories, Stockholm, Sweden, version 9B and version 11) using RayStation’s model -based segmentation algorithm (MBS). Image preprocessing and features extraction were performed on Python (version 3.7.16). Features with near-zero variance and high correlation (Spearman > 0.95) were excluded. The remaining features were clustered by an iterative clustering algorithm, grouping highly correlated features (Spearman > 0.75). Only the feature most associated with the investigated outcome (the lowest p-value from the logistic regression model) was retained. The procedure was iterated until the correlation in finally selected features resulted < of 0.75. A coefficient for each feature was obtained by the logistic Least Absolute Shrinkage and Selection Operator (LASSO) Regression Model. The resulting radiomic score for each patient was calculated by summing the features multiplied by their relative coefficient. Three distinct predictive models were developed: the clinical model — containing only clinical information; the radiomic model — containing only the radiomic score; clinical-radiomic model — containing both clinical information and a radiomic score. Logistic regression models were used to test associations between clinical variables and DLCO abnormality. Odds ratio (OR) estimates were quantified and 95% confidence intervals (CI) were presented. Statistical tests as Wilcoxon rank sum test, Pearson's Chi-squared test, Fisher's exact test were performed in order to evaluate if the difference between the clinical data, spirometry data and cohort characteristics found in the two subpopulations (Normal and Abnormal findings for baseline DLCO) was significant.

Results:

A total of 98 patients (125 nodules) met the inclusion criteria, with a median age at diagnosis of 70 years (IQR 61-79 years). Patients’ baseline characteristics (overall and for the subgroups with Normal/Abnormal baseline DLCO) are shown in Table 1 . Considering pulmonary function, median DLCO before and after radiation was 74 mL/min/mm Hg (59-84) and 68 mL/min/mm Hg (54-76), respectively. Abnormal DLCO values resulted significantly associated with Charlson Comorbidity Index (CCI) (p = .014). Three models for predicting oncological outcome were created: (i) Model 1 based on the features in phase 0 (ii) Model 2 based on the features in phase 50 (iii) Model 3 based on the difference between the two phases 0 and 50. AUC of the developed radiomic and clinical-radiomic model are reported in Figure 1 . AUC value for the clinical model in was 0.65.