This paper investigates the huge potential provided by the deep learning paradigm associated with a convolutional neural network architecture for automatic diagnosis of focal liver lesions. The dataset was provided by the Department of Gastroenterology and Hepatology, "Victor Babes" University of Medicine and Pharmacy, Timisoara and contains a total of 285 contrast-enhanced ultrasound video sequences of five types of lesions. As elements of novelty of our work we could mention the use of 2D-Deep Convolutional Neural Network for implementing an automated diagnosis system which discriminates between an increased number of focal liver lesion types. Our experimental results demonstrated that the proposed principle achieves a competitive performance, 95.71% accuracy, compared to state-of-the-art methods.