3D Volume Segmentation using Texture Analysis

Computer imaging has become very important in the practice of modern medicine.    In medical images, large voxel data sets, containing information about the internal anatomy and physiology of a patient, are obtained from a variety of imaging devices, such as Computer Tomographic (CT scan) or Magnetic Resonance Imaging (MRI).  Segmentation of structures from these volume data sets is a challenging data-dependent task.

Differences in intensity values alone are not adequate to properly segment a volume.   I rely on differences in the spatial arrangement of pixel values neighboring pixels or on the differences in texture.  I use a 3-dimensional bank of Gabor filters to capture the different texture properties of the volume.  A classification algorithm is then used to classify each voxel in the volume.

Image 1: multiple view of input mri data  

Image 2: slice view of input mri data

Image 3: segmented image data. Data segmented into 3
different segments. Lower left (internal fluids and tissues),
lower middle (brain matter) and lower right (skin).

Image 4: slice view of segmented image data