Novel Radiomic Signature as a Prognostic Biomarker for Locally Advanced Rectal Cancer

Background: Locally advanced rectal cancer (LARC) patient stratification by clinicoradiologic factors may yield variable results. Therefore, more efficient prognostic biomarkers are needed for improved risk stratification of LARC patients, personalized treatment, and prognostication. Purpose/Hypothesis: To compare the ability of a radiomic signature to predict disease-free survival (DFS) with that of a clinicoradiologic risk model in individual patients with LARC. Study Type: Retrospective study. Population: In all, 108 consecutive patients (allocated to a training and validation set with a 1: 1 ratio) with LARC treated with neoadjuvant chemoradiotherapy (nCRT) followed by total mesorectal excision (TME). Field Strength/Sequence: Axial 3D LAVA multienhanced MR sequence at 3T. Assessment: ITK-SNAP software was used for manual segmentation of 3D pre-nCRT MR images. All manual tumor segmentations were performed by a gastrointestinal tract radiologist, and validated by a senior radiologist. The clinicoradiologic risk factors with potential prognostic outcomes were identified in univariate analysis based on the Cox regression model for the whole set. The results showed that ypT, ypN, EMVI, and MRF were potential clinicoradiologic risk factors. Interestingly, only ypN and MRF were identified as independent predictors in multivariate analysis based on the Cox regression model. Statistical Tests: A radiomic signature based on 485 3D features was generated using the least absolute shrinkage and selection operator (LASSO) Cox regression model. The association of the radiomic signature with DFS was investigated by Kaplan-Meier survival curves. Survival curves were compared by the log-rank test. Three models were built and assessed for their predictive values, using the Harrell concordance index and integrated time-dependent area under the curve. Results: The novel radiomic signature stratified patients into low-and high-risk groups for DFS in the training set (hazard ratio [HR] =6.83; P < 0.001), and was successfully validated in the validation set (HR=2.92; P < 0.001). The model combining the radiomic signature and clinicoradiologic findings had the best performance (C index=0.788, 95% confidence interval [CI] 0.72-0.86; integrated time-dependent area under the curve of 0.837 at 3 years). Data Conclusion: The novel radiomic signature could be used to predict DFS in patients with LARC. Furthermore, combining this radiomic signature with clinicoradiologic features significantly improved the ability to estimate DFS (P=0.001, 0.005 in training set and in validation set, respectively), and may help guide individualized treatment in such patients.