Pathological automatic classification of hepatocellular carcinoma based on adaptive weighted multi-classifier fusion

In order to assist doctors in predicting the pathological information of hepatocellular carcinoma (HCC) using magnetic resonance imaging (MRI), we proposed an automatic histological grading method of HCC based on adaptive weighted multi-classifier fusion in this paper. First, five sets of texture features were extracted for each region of interest (ROI), corresponding to first order statistics, gray-level co-occurrence matrix, gray-level run-length matrix, Haar wavelet and local binary pattern. Secondly, we selected five single-layer classifiers in the commonly used classifiers for the five feature sets. Thirdly, based on the k-nearest neighbor method and the weighted similarity value, the adaptive dynamic weights of each classifier were calculated. Finally, majority voting procedure was adopted to realize multi-classifier fusion. 192 MRI images of 46 histologically proven HCCs were retrospectively studied. The best classification accuracy (97.06%) was achieved by our method, and the accuracy of different pathological classification were respectively 98.04% (well differentiation) and 96.08% (poorly differentiation). The comparative assessment of the various kind of classification shows that our method, which combining five single-layer classifiers of different types with the adaptive weighted voting rule, is relatively accurate to assist pathological diagnosis of HCCs from enhanced MRI images.