Machine-learning based hybrid-feature analysis for liver cancer classification using fused (MR and CT) images

Abstract

The purpose of this research is to demonstrate the ability of machine-learning (ML) methods for liver cancer classification using a fused dataset of two-dimensional (2D) computed tomography (CT) scans and magnetic resonance imaging (MRI). Datasets of benign (hepatocellular adenoma, hemangioma, cyst) and malignant (hepatocellular carcinoma, hepatoblastoma, metastasis) liver cancer were acquired at Bahawal Victoria Hospital (BVH), Bahawalpur, Pakistan. The final dataset was generated by fusion of 1200 (100 x 6 x 2) MR and CT-scan images, 200 (100 MRI and 100 CT-scan) images size 512 x 512 for each class of cancer. The acquired dataset was preprocessed by employing the Gabor filters to reduce the noise and taking an automated region of interest (ROIs) using an Otsu thresholding-based segmentation approach. The preprocessed dataset was used to acquire 254 hybrid-feature data for each ROI, which is the combination of the histogram, wavelet, co-occurrence, and run-length features, while 10 optimized hybrid features were selected by employing (probability of error plus average correlation) feature selection technique. For classification, we deployed this optimized hybrid-feature dataset to four ML classifiers: multilayer perceptron (MLP), support vector machine (SVM), random forest (RF), and J48, using a ten fold cross-validation method. MLP showed an overall accuracy of (95.78% on MRI and 97.44% on CT). Unfortunately, the obtained results were not promising, and there were some limitations due to the different modalities of the dataset. Thereafter, a fusion of MRI and CT-scan datasets generated the fused optimized hybrid-feature dataset. The MLP has shown a promising accuracy of 99% among all the deployed classifiers.