AUTHOR=Yi Xiaoping , Pei Qian , Zhang Youming , Zhu Hong , Wang Zhongjie , Chen Chen , Li Qingling , Long Xueying , Tan Fengbo , Zhou Zhongyi , Liu Wenxue , Li Chenglong , Zhou Yuan , Song Xiangping , Li Yuqiang , Liao Weihua , Li Xuejun , Sun Lunquan , Pei Haiping , Zee Chishing , Chen Bihong T.
TITLE=MRI-Based Radiomics Predicts Tumor Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer
JOURNAL=Frontiers in Oncology
VOLUME=9
YEAR=2019
URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2019.00552
DOI=10.3389/fonc.2019.00552
ISSN=2234-943X
ABSTRACT=Background: Conventional methods for predicting treatment response to neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC) are limited.
Methods: This study retrospectively recruited 134 LARC patients who underwent standard nCRT followed by total mesorectal excision surgery in our institution. Based on pre-operative axial T2-weighted images, machine learning radiomics was performed. A receiver operating characteristic (ROC) curve was performed to test the efficiencies of the predictive model.
Results: Among the 134 patients, 32 (23.9%) achieved pathological complete response (pCR), 69 (51.5%) achieved a good response, and 91 (67.9%) achieved down-staging. For prediction of pCR, good-response, and down-staging, the predictive model demonstrated high classification efficiencies, with an AUC value of 0.91 (95% CI: 0.83–0.98), 0.90 (95% CI: 0.83–0.97), and 0.93 (95% CI: 0.87–0.98), respectively.
Conclusion: Our machine learning radiomics model showed promise for predicting response to nCRT in patients with LARC. Our predictive model based on the commonly used T2-weighted images on pelvic Magnetic Resonance Imaging (MRI) scans has the potential to be adapted in clinical practice.
Novelty and Impact Statements: Methods for predicting the response of the locally advanced rectal cancer (LARC, T3-4, or N+) to neoadjuvant chemoradiotherapy (nCRT) is lacking. In the present study, we developed a new machine learning radiomics method based on T2-weighted images. As a non-invasive tool, this method facilitates prediction performance effectively. It achieves a satisfactory overall diagnostic accuracy for predicting of pCR, good response, and down-staging show an AUC of 0.908, 0.902, and 0.930 in LARC patients, respectively.