Automated volumetry of meningiomas in contrast-enhanced T1-Weighted MRI using deep learning

Abstract

Background: Meningiomas are among the most common intracranial tumors. In these tumors, volumetric assessment is not only important for planning therapeutic intervention but also for follow-up examination.However, a highly accurate automated volumetric method for meningiomas using single-modality magnetic resonance imaging (MRI) has not yet been reported. Here, we aimed to develop a deep learning-based automated volumetry method for meningiomas in MRI and investigate its accuracy and potential clinical applications.

Methods: For deep learning, we used MRI images of patients with meningioma who were referred to Osaka University Hospital between January 2007 and October 2020. Imaging data of eligible patients were divided into three non-overlapping groups: training, validation, and testing. The model was trained and tested using the leave-oneout cross-validation method. Dice index (DI) and root mean squared percentage error (RMSPE) were measured to evaluate the model accuracy. Result: A total of 178 patients (64.6 ± 12.3 years [standard deviation]; 147 women) were evaluated. Comparison of the deep learning model and manual segmentation revealed a mean DI of 0.923 ± 0.051 for tumor lesions. For total tumor volume, RMSPE was 9.5 ± 1.2%, and Mann-Whitney U test did not show a significant difference between manual and algorithm-based measurement of the tumor volume (p = 0.96).

Conclusion: The automatic tumor volumetry algorithm developed in this study provides a potential volume-based imaging biomarker for tumor evaluation in the field of neuroradiological imaging, which will contribute to the optimization and personalization of treatment for central nervous system tumors in the near future.

Keywords: Algorithm; Biomarkers; Brain neoplasms; Deep learning; Magnetic resonance imaging; Meningeal neoplasms; Meningioma.

Fig. 1

Flowchart depicting the patient selection process.

Fig. 2

(A)Plot of actual vs. predicted tumor volume. Tumor volume was significantly correlated with the predicted volume (R: 0.9877, p: 9.9084e-144; Pearson correlation analysis). (B) Plot of tumor volume vs. Dice index (DI). Tumor volume showed a weak correlation with the DI (R: 0.21283, p: 0.0043411; Pearson correlation analysis).

Fig. 3

Representative cases are shown. The left column shows a contrast-enhanced T1 image with the outline of a detected tumor. The right column image is a magnified image of the tumor and its surrounding area. The blue contour shows manual segmentation, while the red contour represents automated segmentation using the algorithm.A (case no. 30) is a case of left frontal base meningioma with a Dice index (DI) of 0.975, which is representative of highly accurate segmentation of the region bordering the superior sagittal sinus and falx cerebri.B (case no. 212) is a case of right convexity meningioma. Although the tumor was irregularly shaped, it was segmented with high accuracy (DI: 0.973).C (case no. 109) is a case of right parasagittal meningioma with a DI of 0.688 and poor segmentation. The tumor was small, and the contrast effect with the surrounding parenchyma was poor.