3D Volume Reconstruction from MRI Slices based on VTK

Abstract

In today's fast-advancing world, Deep learning brought the huge potential to the healthcare system and it still undergoes different amazing new techniques. New automatic brain tumor segmentation models have been realized. As a result, it is being much more affordable and faster to save lives. However, most of the tumor detection works are still being conducted with 2D single slices of brain image, although, there are new 3D CNN [1] models with more benefits. Those 3D models enable to scan of brain images in 3d volume. 2D models accept only single slices as input and they innately fail to use context from neighboring slices. Missed voxel data from contiguous slices might affect the detection of tumors and decrease the accuracy of the model. 3D models address this issue by utilizing 3D convolutional kernels to make predictions from volumetric inputs. The capacity to use interslice features can increase the further performance of the model. Therefore, in practice, 3D volumes enable to obtain much more efficient and clear diagnoses. İn this paper we purpose our new 3D MRI reconstruction algorithm based on VTK toolkit [3].