Deep learning for automated segmentation of brain edema in meningioma after radiosurgery

Although gamma Knife radiosurgery (GKRS) is commonly used to treat benign brain tumors, such as meningioma, irradiating the surrounding brain tissue can lead to perifocal edema within a few months after the procedure. Volumetric assessment of perifocal edema is crucial for therapy planning and monitoring. Post-radiosurgery changes in perifocal edema, appearing as hyper-dense areas in magnetic resonance T2-weighted (T2w) images, are clearly identifiable; however, physicians lack tools to segment and quantify the volume of these T2w hyper-dense areas. This has hindered not only the quantification of severity but also research on edema growth and case differentiation.

In this study, we trained a Mask Region-based Convolutional Neural Network (Mask R-CNN) to replace manual pre-processing in designating regions of interest. We also applied transfer learning to the DeepMedic deep learning model to facilitate the automatic segmentation and quantification of brain edema regions in images. The resulting quantitative findings were used to explore the effects of GKRS treatment on brain edema caused by meningioma.

We studied 21 patients with meningiomas who had undergone GKRS treatment based on 154 regularly tracked T2w scans. From this group, we selected 130 scans for random assignment to a training set (80 scans), validation set (30 scans), and test set (20 scans). The actual range of the edema in the T2w images was labeled manually by a clinical radiologist to serve as the gold standard in supervised learning. The trained model was tasked with segmenting the test set for comparison with the manual segmentation results. The average Dice similarity coefficient in these comparisons was 84.7%.

The proposed scheme for the automated segmentation and quantification of brain edema post-radiosurgery demonstrated excellent results, suggesting its applicability to the development of predictive models.

Not applicable.