Automatic classification and differentiation of atherosclerotic lesions in swine using IVUS and texture features

Our goal was to develop an automatic classification algorithm to differentiate between four common lesion types in atherosclerotic (AS) arteries: calcific (CAL), fibro-calcific (FBC), fibrous (FBR), and fibro-fatty (FBF). AS was induced in eight Yucatan miniswine. 22 femoral or carotid arteries were imaged with intravascular ultrasound using a pull-back procedure. Both 2D and 3D texture measures were used, followed by a principal components analysis to reduce dimension. The classifiers were applied to the test dataset, and the results were compared with two independent experts. There was no difference between the 2D and 3D classification of the CA and E1, and of the CA and E2 (ANOVA, F = 2.00). The difference between CA and E1 (or E2) was not larger than the difference between E1 and E2 for any lesion type (ANOVA, F = 0.76). We conclude that using 3D information in the classification scheme improved the algorithm's ability to correctly classify lesion type.