A Neuronavigation Toolkit for 3D Visualization, Spatial Registration and Segmentation of Brain Vessels from MR Angiography Images

Neuronavigations are real-time approaches implemented to help neurosurgeons precisely localize different intracerebral pathologies by using multiple neuroimaging modalities. However, current systems suffer from several shortcomings including time-consuming, 3D registration and segmentation accuracies as well as difficulties transferring 3D vessel imaging to the neuronavigation systems. In this work, we introduce a standalone platform for image-fusion, and semantic segmentation of MR angiography images supported for neurovascular interventions. The full-stack toolkit consists of two parts: back-end and front-end. At back-end, Tensorflow library and Ajax for application programming interface server request-response interactions are embedded. As to front-end, we implement C++_based Qt platform for GUI (Graphic User Interface) in Visual Studio integrated development environment. The toolkit provides 3D volume and slice-based visualizations of brain multimodal images. All medical images are archived in DICOM standard, then converted into NIfTI formats. Visualization Toolkit was used to render 3D MR images. The implemented volume-rendering techniques allow the direct visualization of vascular structures, thus reveal vessel abnormalities more faithfully. We provide GUI interface of the modality brain image registration and segmentation functionality. There are two machine-learning based automatic registration categories: (1) intensity-based, (2) geometry-based. The registration error in mm is computed by using Hausdorff Distance metric. As for cerebrovascular segmentation, a method that utilizes an enhanced vesselness filter which bases on multiscale Hessian eigenvalues is performed to extract neurovascular trees. The introduced toolkit is expected to be a validated platform that allows researchers to apply insightful results into the operating room for clinical evaluations.