Texture analysis of human liver

Abstract

Purpose: To classify healthy and diseased livers by texture analysis (TA).

Materials and methods: We studied 43 patients divided into four groups according to their clinical stage and 10 controls on a 1.5-T magnetic resonance (MR) imager, using a T2-weighted breath-hold sequence. For the TA, features of the first and second order were used, and several classification procedures were applied for the classification of patients and controls. The choice of features was performed manually and by use of the Fischer coefficient, average correlation coefficients between features and multidimensional discrimination measure.

Results: All the statistical methods employed were able to differentiate between controls and patients in each group. The classification error varied around 8%.

Conclusion: We have shown that texture analysis can be successfully used for separating cirrhotic patients and healthy volunteers. Different sets of TA features can be used for a similar classification of patients.