Radiomics Based on MR Imaging of Rectal Cancer: Assess Treatment Response to Neoadjuvant Chemoradiotherapy
Fu Shen1, Jie Li2, and Jianping Lu1

1Radiology Department, Changhai Hospital, Shanghai, China, 2Huiying Medical Technology Co., Ltd., Beijing, China


The goal of this study was to investigate the value of high resolution T2-weighted–based radiomics in prediction of treatment response to neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC). The result demonstrated that the MRI based radiomics machine learning model could assess tumoral treatment response to nCRT in patients with LARC.


The preoperative treatment of nCRT in patients with locally advanced rectal cancer (LARC) has resulted in a decrease in the local recurrence rate and an increase in the rates of long-term survival and sphincter preservation. Recent studies have shown that magnetic resonance based radiomics analysis has important values in identifying tumor heterogeneity and can add a further dimension to the predictive power of imaging [1-3]. The purpose of this study was to investigate the imaging response to nCRT by using high resolution T2-weighted based radiomics in patients with LARC.


Forty-six patients with LARC underwent rectal MR imaging before nCRT in our hospital on a 3T scanner (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) between March 2016 and July 2018. Surgical histopathologic tumor regression grade (TRG) was used as the reference standard to evaluate the outcome of the treatment. TRG can be graded as 0, 1, 2, 3, in which grade 0 means no residual tumor cells, grade 1 means single cancer cell or small foci of tumor cells, grade 2 means some tumor cells left and fibrosis are seen, grade 3 means extensive residual tumors and no or little neocrosis. Among all four groups, TRG 0 and TRG 1 were classified as “good” group, and TRG 2 and TRG 3 was “poor” group. One radiologist segmented the lesions on high resolution T2-weighted MRI. One thousand and twenty-nine features were extracted on a radiomics analysis platform (Radcloud, Huiying Medical Technology, Beijing, China), then the features were selected by least absolute shrinkage and selection operator (LASSO) method to best predict the treatment performance. In the LASSO method, leave-one-out cross validation (LOOCV) was used to select the optimal regularization parameter alpha, as the average of mean square error on each patient was smallest. With the optimal alpha, features that had non-zero coefficient in LASSO was reserved. The Logistic Regression (LR) classifier was trained to predict which group the patient belong to. As 5-fold cross validation was used in the experiment, the receiver operating characteristic (ROC) curves for every fold were obtained.


Of the 46 patients, 4 (8.7%) had TRG 0, 15 (32.6%) had TRG 1, 17 (37.0%) had TRG 2, and 10 (21.7%) had TRG 3. One thousand and twenty-nine features were extracted before nCRT, and 28 characteristic features related to the TRG prediction were obtained while the optimal regularization alpha was 0.0267. For the LR classifier, the cross validation received mean area under the ROC curve (AUC) 0.979 (95%CI: 0.959,0.999), mean sensitivity 0.90 and mean specificity 0.933 (Figure 3, 4).


The initial purpose was to help predict TRG of patients with LARC with just MR images before nCRT. High-resolution T2-weighted MRI-based radiomics was used on an attempt to quantify treatment response to nCRT. By using the pretreatment MRI data, we developed a radiomics model with excellent performance for individualized, noninvasive prediction of tumor regression in patients with LARC and potentially guide treatments to select patients for a "wait-and-see" policy.


Our study demonstrated that the high resolution T2-weighted MRI-based radiomics before nCRT showed good classification performance related to TRG in patients with LARC. The radiomics as imaging biomarkers have the potential to prediction of tumoral treatment response to nCRT.




[1] Robert JG, Paul EK, Hedvig H, et al. Radiomics: image are more than pictures, they are data[J]. Radiology. 2016; 278(2):563-577.

[2] Wu J, Tha KK, Xing L, et al. Radiomics and radiogenomics for precision radiotherapy[J]. J Radiat Res. 2018; 59(suppl_1): i25-i31.

[3] Horvat N, Veeraraghavan H, Khan M, et al. MR Imaging of Rectal Cancer: Radiomics Analysis to Assess Treatment Response after Neoadjuvant Therapy[J]. Radiology. 2018; 287(3):833-843.


Mean square error on each patient.

Coefficient for each feature

The ROC curve for LR classifier

Cross validation result.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)