Inter-Fractional Spatiotemporal Dosiomics Improves the Prediction of Local Progression for Patients with Small Cell Lung Cancer
Abstract
Purpose
To propose and validate a novel dosimetric method integrating inter-fractional temporal dose changes for improved SCLC prognosis management and individualized decision-making.
Methods
The 144 patients with small cell lung cancer (SCLC) were split into training and test sets (8:2) for 5-fold cross-validation. Planned CT (pCT) was deformed to cone beam CT (CBCT) for registered CT (rCT) generation and dose recalculation to quantify the inter-fractional varieties. Dosiomics features were extracted from the dose distributions of each fraction. The delta-dosiomics was defined as the discrepancies of dosiomics features on the pCT and rCT of various fractions. The spatiotemporal dosiomics was constructed by stacking the fractional differences. After feature selection and dimension reduction using principal component analysis (PCA), the performance of the temporal dosiomics model was evaluated on the test set. Kaplan-Meier and log-rank tests assessed survival.
Results
The Area Under the Curve (AUC) and Area Under the Precision-Recall Curve (PR_AUC) of the proposed temporal model were significantly superior to that of the conventional approach based on the planned dose distribution (AUC: 0.81 vs. 0.73, PR_AUC: 0.76 vs. 0.65). Kaplan-Meier analysis revealed that the survival curves of the high-risk and low-risk groups as classified by both models were significantly separated (p<0.001). However, the hazard ratio (HR=0.222, 95%CI: 0.125-0.392) of the survival curve from the spatiotemporal dosiomics model was 24% lower than that from the traditional dosiomics model (HR=0.291, 95%CI: 0.158-0.533).
Conclusion
Incorporating inter-fractional dosimetric variations enhances SCLC local progression prediction accuracy, facilitating precision radiotherapy and personalized follow-up.