Purpose: To fully automate intra-abdominal (IAT) and total adipose tissue (TAT) segmentation in mice to replace tedious and subjective manual segmentation.
Materials and methods: A novel transform codes each voxel with the radius of the narrowest passage on the widest possible three-dimensional (3D) path to any voxel in the target object to select appropriate IAT seed points. Then competitive region growing is performed on a distance transform of the fat mask such that competing classes meet at narrow passages effectively segmenting the IAT and subcutaneous adipose compartments. Fully automatic segmentations were conducted on 32 3D mouse images independent to those used for algorithm development.
Results: Automatic processing worked on all 32 images and took 28 s on a 3.6 GHz Pentium computer with 2.0 GB RAM. Manual segmentation by an experienced operator typically took 1 h per 3D image. The correlation coefficients between manual and automated segmentation of TAT and IAT were 0.97 and 0.94, respectively.
Conclusion: The fully automatic method correlates well with manual segmentation and dramatically speeds up segmentation allowing MRI to be used in the anti-obesity drug discovery pipeline.