| Abstract: |
Background/Objectives: Delta-radiomics involves analyzing feature variations at different acquisition time-points. This study aimed to assess the utility of delta-radiomics feature analysis applied to contrast-enhanced (CE) and non-contrast-enhanced (NE) T1-weighted images (WI) in predicting the therapeutic response to chemoradiotherapy (CRT) in patients diagnosed with muscle-invasive bladder cancer (MIBC). Methods: Forty-three patients with non-metastatic MIBC (cT2–4N0M0) who underwent partial or radical cystectomy after induction CRT were, retrospectively, reviewed. Pathological complete response (pCR) to CRT was defined as the absence of residual viable tumor cells in the cystectomy specimen. Identical volumes of interest corresponding to the index bladder cancer lesions on CE- and NE-T1WI on pre-therapeutic 1.5-T MRI were collaboratively delineated by one radiologist and one urologist. Texture analysis was performed using “LIFEx” software. The subtraction of radiological features between CE- and NE-T1WI yielded 112 delta-radiomics features, which were utilized in multiple machine-learning algorithms to construct optimal predictive models for CRT responsiveness. Additionally, the predictive performance of the radiomics model constructed using CE-T1WI alone was assessed. Results: Twenty-one patients (49%) achieved pCR. The best-performing delta-radiomics model, employing the “Extreme Gradient Boosting” algorithm, yielded an area under the receiver operating characteristic curve (AUC) of 0.85 (95% confidence interval [CI]: 0.75–0.95), utilizing four signal intensity-based delta-radiomics features. This outperformed the best model derived from CE-T1WI alone (AUC: 0.63, 95% CI: 0.50–0.75), which incorporated two morphological features and one signal intensity-based radiomics feature. Conclusions: Delta-radiomics analysis applied to pre-therapeutic CE- and NE-MRI demonstrated promising predictive ability for CRT responsiveness prior to treatment initiation. © 2025 by the authors. |
