User-Guided Semi-Automatic Segmentation of Medical 3D Images

A research project by Sander Vanheste
Conducted at Utrecht University at the Master program
Under supervision of Amir Vaxman, K. Vincken, C. Mol
Apr 2019

In cooperation with the Utrecht Medical Center (UMC), I built a 3D browser-based MRI/CT e-learning system. I propose an improved way of defi ning correct target volumes in 3D MRI and CT scans: instead of user-defi ned, handcrafted target volumes in which the target contour of a specfii c anatomical structure has to be traced by hand for each 2D image slice, this method introduces automatic interpolation between user-provided contours. This is achieved with a new variant of the active contour method. As a result, only a subset of the slices have to be segmented manually. This improves on similar contour-based user-guided segmentation methods.

segmentation;active contours;medical e-learning;interpolation;3d-segmentation;medical datasets