Prediction of recurrence risk of cervical cancer after radiotherapy using multi-sequence MRI radiomics

Abstract

Objective

To predict the recurrence risk of cervical cancer after radiotherapy using multi-sequence magnetic resonance imaging (MRI) radiomics.

Methods

A total of 90 cervical cancer patients treated in the First Affiliated Hospital of Soochow University from January 2018 to January 2023 were enrolled in this retrospective study, comprising 29 cases with recurrence and 61 cases without recurrence. The cohort was divided into a training set of 60 cases and a test set of 30 cases. Tumor regions of interest (ROI) were delineated using MRI radiomics scans before and after treatment, and image features were extracted to build predictive models. Ten models were used to predict recurrence risk in the test set, named as combined model T1-weighted imaging (T1WI) sequence, combined model fast gradient-recalled echo (FGRE) sequence, combined model T2 fat suppression sequence, combined model-epi sequence, FGRE sequence-T1WI sequence model, FGRE sequence-T2 fat suppression sequence, FGRE sequence-epi sequence model, T2 fat suppression sequence-T1WI sequence model, T2 fat suppression sequence-epi sequence model and the combined multi-sequence model.

Results

In the training set, compared with the combined multi-sequence model, the receiver operating characteristic (ROC) curves of the T1WI sequence, FGRE sequence, and T2 fat suppression sequence combined with the T1WI sequence model were significantly different (Z ​= ​2.25, 2.66,2.54, P ​< ​0.05). In the test set, the ROC curve of the T1WI sequence model also showed a statistically significant difference from the combined model (Z ​= ​2.21, P < ​0.05). The T1WI sequence, FGRE sequence, T2 fat suppression sequence, EPI sequence, and the combined model were all effective in predicting post-radiotherapy cervical cancer recurrence [area under curve (AUC) ​= ​0.731, 0.705, 0.823, 0.754, 0.871, P ​< ​0.05]. Compared with the single-sequence models, the combined multi-sequence model showed the highest AUC value, accuracy, and precision in the ROC curve (AUC ​= ​0.854, P ​< ​0.05).

Conclusion

Multi-sequence MRI radiomics could effectively predict the risk of cervical cancer recurrence after radiotherapy, and the combined multi-sequence model demonstrates enhanced predictive performance.