Development of a Visual Assessment System for Meningiomas Using Deep-Learning Technology: A Multicenter Study

Background: Accurate assessment is important for meningioma patients in magnetic resonance imaging (MRI). The purpose of this study was to establish an automated visual evaluation system to facilitate diagnosis and treatment. Methods: The system, designed with a cascade network structure, was developed with deep-learning technology for automatic function of tumor detection, visual assessment, and grading prediction. Patients were retrospectively collected from two institutions. Deep-learning models were trained on the largest meningioma dataset so far involved 97500 MR images of 625 patients from the first institution. Specifically, a convolutional neural network model was established first to segment the tumor images, followed by rendering algorithms for spatial reconstruction. Then, another model was trained with the segmented tumor images for grading prediction. After training and validating performances in the first dataset, we integrated our models as a system, and tested the robustness based on performances on the second dataset from the second institution involving different MRI platforms. Findings: The segment model represented worthy-noted performances with Dice coefficients of 0.920±0.009, and the classification model also achieved high accuracy with AUC of 0.918±0.006 and accuracy of 0.901±0.039 when classifying the tumors into low-grade and high-grade meningiomas. The system showed clinical potentials, and represented good performances in the external validation group. Interpretation: Deep-learning based system could potentially serve as a reliable assessment for meningioma patients. Funding Statement: This work was supported by 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University; Key research and development project of science and technology department of Sichuan Province. Declaration of Interests: None declared. Ethics Approval Statement: Committee of Sichuan University have given approval for statistics export and utilization for this study. The obligatory written informed consent was obtained from participants enrolled in this study (written informed consent for patients <16 years old was signed by parents or guardians).