Effect of Slice Thickness on Texture-Based Classification of Liver Dynamic CT Scans

This paper assesses the impact of slice thickness on texture parameters. Experiments are performed on liver dynamic CT scans, with two slice thicknesses. Three acquisition moments are considered: without contrast, in arterial and in portal phase. In total, 155 texture parameters, extracted with 9 methods, are tested. Classification of normal and cirrhotic liver is performed using a boosting algorithm. Experiments reveal that slice thickness does not considerably influence the stability of the parameters. They also enable to assess the rate of parameter dependency on slice thickness. Finally, they show that applying different slice thicknesses for training and testing the CAD system requires slice thickness-independent parameters.