HCC Recognition Within Ultrasound Images Employing Advanced Textural Features with Deep Learning Techniques

HCC (Hepatocellular carcinoma) is the liver cancer form that is the most often met. The actual standard for a highly accurate diagnosis of HCC is the biopsy, this being, however, invasive, as well as dangerous. We develop computerized methods, based on advanced image processing, in order to perform automatic, respectively computer aided diagnosis of HCC based on ultrasound images. In this research paper, we analyze the role of several advanced texture analysis specific methods, combined with powerful conventional classifiers, respectively deep learning techniques, for HCC automatic recognition. Regarding the deep learning methods, we focus on Stacked Denoising Autoencoders (SAE), so we study its impact upon the performance of HCC computerized diagnosis. In our experiments, we compare the HCC tumor with both the cirrhotic liver parenchyma on which it evolved and with the hemangioma tumor, the obtained accuracy being around 85%. The actual results are compared, as well, with the previously obtained results in the literature.