An automated approach for predicting glioma grade and survival of LGG patients using CNN and radiomics

• DOI: 10.3389/fonc.2022.969907
• Publication Date: 2022-08-12T11:29:30Z
• Authors: Chenan Xu; Yuanyuan Peng; Weifang Zhu; Zhongyue Chen; Jianrui Li; Wenhao Tan; Zhiqiang Zhang; Xinjian Chen
• Keywords: 03 medical and health sciences; 0302 clinical medicine; Oncology; glioma grade; radiomics; convolutional neural network; Neoplasms. Tumors. Oncology. Including cancer and carcinogens; survival of lower-grade glioma; RC254-282; automatic diagnosis; 3. Good health

To develop and validate an efficient automatically computational approach for stratifying glioma grades predicting survival of lower-grade (LGG) patients using integration state-of-the-art convolutional neural network (CNN) radiomics. This retrospective study reviewed 470 preoperative MR images from BraTs public dataset (n=269) Jinling hospital (n=201). A fully automated pipeline incorporating tumor segmentation grading was developed, which can avoid variability subjectivity manual segmentations. First, integrated by fusing CNN features radiomics employed to stratify grades. Then, a deep-radiomics signature based on the LGG developed subsequently validated in independent cohort. The performance achieved Dice coefficient 0.81. intraclass correlation coefficients (ICCs) between physicians were all over 0.75. area under curve (AUC) 0.958, showing effectiveness approach. multivariable Cox regression results demonstrated that remained prognostic factor nomogram showed significantly better than clinical overall (C-index: 0.865 vs. 0.796, P=0.005). proposed be noninvasively efficiently applied prediction gliomas grade survival. Moreover, our successfully computerized instead segmentation, shows potential reproducible practice.