Dosimetry-guided morphomic modeling using daily FBCT for automated all-in-one radiotherapy eligibility assessment in postoperative cervical cancer

Objective This study aimed to develop an automated decision-making model based on daily fan-beam CT (FBCT) morphomic features to determine whether a treatment fraction is suitable for AIO-RT initiation. Methods A total of 48 postoperative cervical cancer patients with 313 FBCT fractions were retrospectively analyzed. Morphomic features were extracted from bladder, rectum, intrarectal gas, bilateral femoral head, and inter-organ spatial relationships. Feature selection was performed using least absolute shrinkage and selection operator (LASSO), and predictive models were built using Random Forest (RF) and Gradient Boosting Machine (GBM) under leave-one-patient-out cross-validation. Results Both models achieved consistent discriminative performance (AUC = 0.958 for RF and 0.952 for GBM). Bladder-related morphological features were the dominant predictors, particularly bladder volume and shape-derived descriptors. Spatial features also contributed to classification performance, while rectal gas showed limited but complementary predictive value. Conclusion The proposed FBCT-based morphomic framework can accurately identify unsuitable fractions for AIO-RT before workflow initiation, providing a practical and interpretable decision-support tool for adaptive radiotherapy workflow optimization.