ESTRO 2025 - Abstract Book

S3747

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Results: Results are shown in Table 1 and Figure 1. Model 1, based on Statistical-Percentile10, Morphological-ComShift and stage, was confirmed in both training (p-value=0.0002; C-index=0.632) and validation groups (p-value=0.053; C index=0.590). Since Statistical-Percentile10 was the primary RF driving this model, considering it alone improved results (Model 2: training: p-value=0.0004; C-index=0.619; validation: p-value=0.0396; C-index=0.601). Finally, a new two-variable model (Model 3), based on Statistical-Percentile10 and GLSZM3D-grayLevelVariance (anti correlated with COMshift) was identified, performing well in both training (p=0.0001; C-index=0.654) and validation (p=0.0072; C-index=0.623). The three models’ performances in terms of C-index were similar (0.590-0.623) in the validation cohort.