Preliminary study on the prediction of acute hematologic toxicity in cervical cancer patients after radiotherapy alone based on CT radiomics

s: Objective This study aims to construct a predictive model for acute hematologic toxicity (HT) in cervical cancer patients undergoing radiotherapy alone by utilizing CT radiomic features prior to treatment, combined with clinical and dosimetric characteristics.Methods We retrospectively analyzed data from 82 cervical cancer patients treated with radiotherapy alone at Affiliated Hospital of Jiangnan University (June 2022–March 2025). Patients were divided into training and test sets (8: 2 ratio), with ≥grade 2 HT as the endpoint. Clinical features and dosimetric parameters of pelvic structures (lumbosacral spine, lower pelvis, ilium, and overall pelvis) were collected. A total of 1, 046 radiomic features were extracted from the entire pelvic structure using PyRadiomics. First, a Clinical-dose model was constructed based on 56 clinical/dosimetric features after Z-score normalization, Spearman correlation analysis (>0.9), and LASSO regression with K-nearest neighbor (KNN) machine learning via 5-fold cross-validation. A Radiomics model was then developed using the same methodology. Finally, a Hybrid model integrating both feature sets was established. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC).Results Of the patients, 48.8% (40/82) developed ≥grade 2 HT. Univariate analysis found that V20 in the lumbosacral spine was significantly associated with the development of grade ≥2 HT (P<0.05). The radiomics prediction model achieved an AUC of 0.802 (95% CI: 0.702-0.902) in the training set and 0.703 (95% CI: 0.425-0.981) in the test set15. The clinical-dose model demonstrated AUC values of 0.750 (95% CI: 0.634-0.866) in the training set and 0.633 (95% CI: 0.362-0.904) in the test set25. The hybrid model outperformed both, with AUCs of 0.861 (95% CI: 0.778-0.944) in the training set and 0.781 (95% CI: 0.550-1.000) in the test set.Conclusion Pre-radiotherapy CT radiomic features combined with dosimetric characteristics can predict the occurrence of acute hematologic toxicity in cervical cancer patients, potentially aiding in early clinical intervention.