Carotid ultrasound image analysis using artificial neural networks

Abstract

This paper aims at developing an ultrasound-based diagnostic measure quantifying plaque activity and the likelihood of asymptomatic lesions to produce neurological symptoms. Based on echogenicity the methodology has been successfully applied on longitudinal ultrasound images of the carotid artery bifurcation. Transverse ultrasound images incorporate noise, artifacts, shadowing and reverberation. Nevertheless, transverse images are a resource not yet fully explored. The comparison of sequential transverse images minimizes the intrinsic scale variability between operators and ultrasound devices. Based on pixel level tissue classification, the use of an artificial neural network analysis appled to transverse images allows identifying vulnerable or unstable echolucent plaques.