A novel clinical decision support system for liver fibrosis using evolutionary multi-objective method based numerical association analysis

Abstract

Chronic liver diseases are among the major health problems in the world. Determining the degree of fibrosis and structural changes and early diagnosis is an important indicator for the course of chronic liver disease, screening of complications and response to treatment. Considering the prevalence of the disease, the use of an invasive biopsy method does not seem practical. At least a preliminary assessment should be able to determine which patients should have a biopsy. In addition, it is not possible to repeat the liver biopsy frequently to follow the course of the patients. Liver biopsy is expensive and it cannot be performed in every hospital. Difficulties in the application for physicians and patients, sampling errors, differences in evaluation, the requirement of a trained physician, difficulties to repeat, and serious complications during the procedure are other disadvantages. The association rule discovery aims to find interesting and valuable associations within the data. Although association analysis is a very useful and popular task in data mining, as far as we know, there is not any study about association analysis of liver fibrosis. We hypothesize at this work that, evolutionary multi-objective methods can be very efficiently modeled and adapted for the automatic miner of comprehensible, accurate, and interesting numerical positive and negative association rules in liver fibrosis clinical decision making. Due to the numerical valued attributes in liver fibrosis data, for the first time, evolutionary intelligent MOPNAR was handled as a rule miner from liver fibrosis without using any discretizing process that requires domain experts. The algorithms modeled for a clinical decision support system in this study modify and adapt themselves for automatic discovery of numerical association rules and do not require modifying or changing the data. Sensitivity analysis of MOPNAR for liver fibrosis was also performed for the first time and a better parameter setting for this task was presented. According to the discovered rules in liver fibrosis data, the MOPNAR outperformed the compared method with respect to average confidence, lift, certainty factor, netconf, yulesQ, number of attributes, and number of covered records.

Keywords: Artificial intelligence; Liver fibrosis; Optimization.