To compare the predictive value of the radiomics siganature extrated from MRI plain or enhancement imaging for the survival of rectal cancer / 中华放射学杂志

ABSTRACT

Objective To compare the predictive value of radiomics signature extracted from MRI plain and enhancement sequence for the disease-free survival (DFS) of rectal cancer. Methods We retrospectively analyzed fifty-one patients with rectal adenocarcinoma confirmed by biopsy from October 2010 to December 2013 in Cancer Hospital Chinese Academy of Medical Sciences.All patients underwent neoadjuvant chemotherapy(nCRT)followed total mesorectal excision(TME),and MRI scans were performed before nCRT.Follow-up time for the survival patients were more than 3 years.The image segmentation was performed on the T2WI sequence of the small FOV and the multi-phase enhancement sequence venous phase,respectively.Least absolute shrinkage and selection operator(LASSO)Cox regression was applied to extract radiomics features and the imaging signature was constructed. According to the radiomics score of each patient,the patients were divided into the high risk group with shorter DFS and the low risk group with longer DFS. A 3-year DFS was calculated for radiomics signature using the Kaplan-Meier product limit method with univariate log-rank analysis testing for differences in the training and validation cohort, respectively. And the predictive ability of the model was evaluated by concordance index (C-index). Results The training set and the validation set were 36 and 15 cases, respectively. During follow-up 32 patients experienced relapse(26 distant,3 local and 3 both),and 19 cases were censored.Twelve features were extracted in the enhanced sequence.The radiomics signatures were significant for DFS in the training set and the validation set(P=0.000 2 and 0.009 1,respectively).The C-index of the model were 0.904 and 0.700 in the training set and the validation set, respectively. The model has the better ability to predict survival.Two features were extracted in the plain sequence.The radiomic signature was significant for DFS in the training set(P=0.005 0),while the radiomics signature was not significant for DFS in the validation set (P=0.767 0). The C-index of the model were 0.711 and 0.500 in the training set and the validation set, respectively.Conclusions Radiomics signature extracted from MRI venous phase enhancement sequence superior to plain sequence for predicting the DFS of rectal cancer before nCRT.