Comparative Neural Network Based Venous Thrombosis Echogenicity and Echostructure Characterization using Ultrasound Images

Venous thrombosis is a common pathology that creates serious public health problems. Thrombosis diagnosis, particularly the determination of their echogenicity and echostructure can be efficiently accomplished by a medical expert using ultrasound imaging. On the other hand, the predictive capability of artificial neural networks is very useful in medical applications and can support medical experts to take appropriate diagnosis decisions. Therefore, the proposed study intends to characterize by means of neural networks the thrombosis echogenicity and echostructure, using a predefined learning base that depends on the prior knowledge of physicians. We have studied six different methods to characterize the thrombosis images, along with the six corresponding neural networks. Obtained results show that the optimal feature vector size, the simplest neural network architecture, and the smallest error, are achieved by using the mean-variance approach or by the wavelet coefficients energies method