Application of segmentation of intravascular images for tissue characterization of vascular pathology

Intravascular images from patients undergoing coronary angioplasty were obtained by a 20 MHz catheter probe. Texture analysis nas performed computing features of different regions of interest, representing son and calcified plaque and thrombus. For each class about 100 feature sets were disposed. computed in regions selected in 30 images. Texture features were classified using Bayesian classifier and a neural back propagation network. The statistical classifier led to a good discrimination between sort and calcified plaque whereas half of the thrombus feature sets were recognized as sort plaque. The accuracy of the classification result when using the neural network classifier was 87% for calcified plaque, 88% for sort plaque and 76% for thrombus. The neural classification process was implemented as a visualization routine for PC supported classification. For this purpose the 51 texture parameters were calculated and sent to the recall routine which delivered the neural network classification result. The classification result were color encoded with red, blue and green labels for calcified plaque, soft plaque and thrombus.