Using MRI data segmentation to deliver a 3D model that can be presented in mixed reality techniques – key difficulties and conclusion in regard to cerebral vessel imaging – case study
November 17, 2022
Pulanecki T, Ozga P, Proniewska K
Mixed reality techniques, such as holographic vision can be used to better visualize medical images, whether for pre-operative/clinical assessment or educational purposes. The possibility of interaction with a 3D model located in mixed reality instead of an usual 2D MRI scan slices shown on a screen, allows the user to better understand the shape, proportions and relations in space between presented objects, which can help a surgeon plan the operation or be an aid for students or patients to clearly visualize and understand the anatomy/pathology. In order to be able to create a model that can be shown in mixed reality, the 2D MRI data needs to be carefully processed and whether this process can be done yielding an valuable, high quality model, depends on a number of factors. The most important one of those is the quality of source data. In this case study, an anonymized DICOM dataset has been processed using an open-source software (3D Slicer v4.11). After analysis of 14 MRI series, the final source series consisted of 170 images, 640x640px in size, with acceptable contrast between tissues. After adjusting the Hounsfield window in Volumes module, a segment has been created in the Segment Editor, with Threshold tool being used in order to try and separate the cerebral vessels from the rest of the image. Due to the nature of the data, there has been no threshold value that would separate the arteries from a number of high-intensity points located mainly in the white matter and skull bone marrow. Using the fact that vessels are continuous and form connections with each other (no artifacts disrupting the continuity of any major vessels have been found in this MRI), the “Islands” function has been used on the model. Using “Keep largest island” algorithm would lead to elimination of the posterior cerebral circulation in this case, due to complete lack of posterior communicating arteries. Instead, after few adjustments, “Remove small islands” allowed the 3D model to consist of 4 segments of connected voxels (Islands). Two of these have been main vessels of anterior and posterior cerebral circulation respectively, and remaining two have been removed manually using “Scissors” function. Because structures of the body in general have a smooth surface, a good practice is to use Smoothing before finishing the model. This tool needs to be used with extreme precision when working on vessel models, both in regard to the algorithm (using opening or closing algorithm could lead to dangerous artifacts, gaussian and joint smoothing would destroy the image) and kernel size. Using median algorithm with kernel size 1mm allows for satisfactory smoothing of the surface without major information loss or artifact creation in this case. As the model is ready, it is exported as a.STL file that can be easily shown in mixed reality setups to help understand the condition it presents. This format would also allow 3D printing of the model. In conclusion – using a simple MRI dataset and open-source software, despite difficulties, a 3D.stl model can be prepared to be presented in mixed reality for better understanding of complex cases regarding cerebral vessels.