Hepatocellular Carcinoma Recognition from Ultrasound Images by Fusing Convolutional Neural Networks at Decision Level

The Hepatocellular Carcinoma (HCC) is the most often met malignant tumor of the liver. It develops from cirrhosis, after a parenchyma restructuring phase, at the end of which dysplastic nodules result that can transform into HCC. Nowadays, HCC diagnosis is usually performed through needle biopsy, also through medical imaging, but most of the methods are invasive and/or expensive. Ultrasonography is the best option for screening, while computerized techniques must be developed as well to derive subtle information. In our previous research, we experimented with conventional techniques, based on advanced texture analysis methods and traditional classifiers, as well as with deep learning techniques, aiming to perform HCC recognition with maximum accuracy. We combined the CNN architectures with each other, respectively with the conventional techniques, at classifier level. In this work, we combined existing and original CNN architectures at decision level, we assessed the corresponding performance, and we compared it with our previous results.