Stage Identification of Malignant Liver Tumors Using Multiclass Classifiers

The work here focuses on the stage-wise classification of liver tumors identified as malignant based on appropriate features. Liver tumors normally progress from stage one to stage four, which indicates the severity of the disease. It is very much essential to identify the stage of cancer to proceed with the treatment by domain experts. In this work, to carry out the stage identification process after recognizing malignant tumors present in the liver, Histogram of Oriented Gradients (HOG) features have been selected and extracted. Prior to this, segmentation of tumors from a liver, validation of tumors and classification of tumors as benign or malignant was carried out. Out of 2640 scans, 1384 scans were classified as containing malignant tumors, and these tumor scans were fed as input to the multiclass classification algorithms such as Random Forest, Naïve Bayes, K-Nearest Neighbor and Convolutional Neural Network to identify stages. The performance of all these classifiers is measured with parameters such as Accuracy, Precision, Recall and F1-Score. It is observed that the CNN has performed better with respect to all parameters. The results obtained are discussed and tabulated.