The Contrast-Enhanced MRI Could Be Substituted by Unenhanced MRI in the Differential Diagnosis of Nasopharyngeal Carcinoma But Not in Segmentation with the Aid of Deep Learning Models : An Exploratory Study in Large-Scale Population of Endemic Area

Abstract

Background: No consensus in sequence selection for artificial intelligence model development has been achieved, we aimed to explore whether contrast-enhanced magnetic resonance imaging (ceMRI) could be substituted in the identification and segmentation of nasopharyngeal carcinoma (NPC) with the aid of deep learning models in a large-scale cohort. Methods: A total of 4,478 eligible individuals were randomly split into training, validation and test sets. The diagnostic performance between NPC and benign hyperplasia and segmentation efficacy in NPC were compared among self-constrained 3D DenseNet models developed using axial T1-weighted imaging (T1WI), T2WI or enhanced T1WI (T1WIC) images separately. Findings: All developed models exhibited similar satisfactory diagnostic performance in discriminating NPC from benign hyperplasia, attaining over accuracy over 99.00% in all T stages of NPC. However, the T1WIC model yielded a significantly higher dice similarity coefficient (DSC) and lower average surface distance (ASD) than either the T1WI model or T2WI model in the test set of NPC cohort (DSC, 0.768±0.070 vs 0.759±0.065 and 0.755±0.071, p < 0.001; ASD, 1.573±0.954 mm vs 1.661±0.898 mm and 1.722±1.133 mm, p < 0.01).Interpretation: Unenhanced MRI could substitute ceMRI in discriminating NPC from benign hyperplasia but not in the segmentation for NPC with the aid of deep learning models and ceMRI remained the optimal sequence for developing segmentation models for NPC , indicating the feasibility of reducing the use of MR contrast in screening program. Funding: This work was supported by the National Natural Science Foundation of China [Grant No. 31900461, 81702873, 81872375, 81572665, 81802712]. Declaration of Interest: Nothing to disclose. Ethical Approval: The ethics committee of the authors’ institution approved the study protocol and patient consent was waived due to the retrospective nature of the study.