Click here to


Are you sure ?

Yes, do it No, cancel

Rectal Cancer Prognosis Prediction Using Radiomics From Pretreatment MRI

X Zhong1*, N Li2 , K Sung1 , X Qi1 , (1) UCLA School of Medicine, Los Angeles, CA, (2) Cancer Hospital Chinese Academy of Medical Sciences, Beijing, China


(Sunday, 7/29/2018) 2:05 PM - 3:00 PM

Room: Davidson Ballroom B

Purpose: To identify significant biomarkers for treatment response prediction before chemoradiation treatment (CRT) for locally advanced rectal cancer (LARC).

Methods: 41 patients enrolled in a prospective trial in a single institution were included. All patients underwent neo-adjuvant CRT with prescription dose of 50 Gy in 25 fractions, followed by total mesorectal excision surgery after completion of the CRT (6-8 weeks). The patients were stratified into responder and non-responder subgroups based on the response assessed by post-operative pathology, MRI or colonoscopy. Prior to the CRT, T2 weighted TSE images were acquired for each patient. The gross tumor volume (GTV) were segmented by the same experienced oncologist on the T2W images. 64 imaging features based on Gray Level Co-Occurrence Matrix (GLCM), intensity statistics as well as Gray Level Run Length Matrix (GLRLM) were extracted from each patient using the IBEX platform. Based on the qualitative property of MRI, the feature extraction was performed based on the dynamic range of the image intensity. The features were selected using both the Wilcoxon rank sum tests with criterion of p < 0.05 between two subgroups and least absolute shrinkage and selection operator (LASSO). The feature selection was performed on each fold during cross validation. The model was evaluated with repeated 10 times five-fold cross validation to reduce the underestimation of variance.

Results: Based on feature selection process intensity skewness, kurtosis, short run low gray level emphasis (SRLGLE) and long run high gray level emphasis (LRHGLE) were shown to be the most important features in the model with an AUC of 0.73 ± 0.15 using LASSO.

Conclusion: This study indicates the feasibility using pre-treatment T2W image to predict the prognosis of the treatment for LARC. The identified imaging features are potential biomarkers for prognosis prediction and individualized therapy.


ROC Analysis, Feature Selection, Feature Extraction


IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics

Contact Email