Patients classification of liver cancer by applying artificial neural networks

Abstract

        The objective of this research is to create a model of liver cancer classification using Artificial Neural Network (ANN) for supporting physicians in the primary diagnosis of liver cancer classification. In this work, the experimental data were obtained from patient registration section and Clinical Laboratory section, Srinagarind Hospital, Faculty of Medicine, Khon Kaen University. This research studied patients who have risk of liver cancer and had been physical examination from Srinagarind Hospital from 1 January 2010 to 31 December 2011. Data were cleaned and transformed into the appropriate format. There were totally 6,499 patient records for experiment. The 3,412 cases are considered as liver cancer patients. Eight relevant variables, including age, gender, occupation, albumin, direct bilirubin, aspartate aminotransferase, alkaline phosphatase, and alanine aminotransferase, were used to construct a classification model by ANN. In this work, the 18 ANN structures, identified by the number of nodes in hidden layer, with various ratios of training, test and validation sets were evaluated to achieve the ANN model structure with highest classification accuracy. The ANN model construction was executed by MATLAB programming. The results showed that classification performance was highest when the number of nodes in hidden layer was 5, and the suitable ratio of the training set, testing set and validation set was 70:20:10. The accuracy of the test set was 95.39% and the overall average of accuracy 94.91%.The obtained Artificial Neural Network model in this research had high classification accuracy. The ANN model could be applied to support physicians in the primary diagnosis of liver cancer classification. Furthermore, the satisfaction evaluation of the ANN model was made by the user group. The evaluation results indicated that the model was manually simple and easy to understand.