Deep learning for locally advanced nasopharyngeal carcinoma prognostication based on pre- and post-treatment MRI

• DOI: 10.1016/j.cmpb.2022.106785
• Publication Date: 2022-03-31T08:58:13Z
• Authors: Song Li; Yu-Qin Deng; Hong-Li Hua; Sheng-Lan Li; Xi-Xiang Chen; Bao-Jun Xie; Zhiling Zhu; Ruoyun Liu; Jin Huang; Ze-Zhang Tao
• Keywords: Concordance; Confusion; Confusion matrix

We aimed to predict the prognosis of advanced nasopharyngeal carcinoma (stage Ⅲ-Ⅳa) using Pre- and Post-treatment MR images based on deep learning (DL).A total 206 patients with primary who were diagnosed treated at Renmin Hospital Wuhan University between June 2012 January 2018 retrospectively selected. A rectangular region interest (ROI), which included tumor area, surrounding tissues organs, was delineated each image. Two Inception-Resnet-V2 transfer models, named Pre-model Post-model, trained Pre-treatment images, respectively. In addition, an ensemble model Post-models established. The three established models evaluated by receiver operating characteristic curve (ROC), confusion matrix, Harrell's concordance indices (C-index). High-risk-related gradient-weighted class activation mapping (Grad-CAM) developed according DL models.The Pre-model, displayed a C-index 0.717 (95% CI: 0.639 0.795), 0.811 0.745-0.877), 0.830 0.767-0.893), AUC 0.741 0.584-0.900), 0.806 0.670-0.942), 0.842 0.718-0.967) for test cohort, comparison performance Post-model better than indicated importance prediction. All performed TNM staging system (0.723, 95% 0.567-0.879). captured features presented Grad-CAM suggested that areas around lymph nodes related tumor.The have staging. are great significance prediction could contribute clinical decision-making.