5510
Somaya Hassan Hashem Ahmed
Liver Fibrosis Stages Classification Using Machine Learning Techniques
Liver Fibrosis, Machine Learning Techniques, Classification Models,Hepatocellular Carcinoma, Particle Swarm Optimization, Genetic Algorithm, Decision Learning Tree.
Artificial Intelligence is today’s language. It is involved in solving various problems. Clinical decision support system establishes a major topic in artificial intelligence. It is a decision support system that focuses on using artificial intelligence according to knowledge management to accomplish clinical advice for patient care based on multiple items of patient data. Chronic hepatitis C (CHC) is recognized as a major healthcare problem worldwide, and as a common infection in Egypt, especially genotype 4. The assessment of liver fibrosis stage in patients with CHC is mandatory for monitoring the prognosis of the disease. Chronic hepatitis has a risk factor for the development of hepatocellular carcinoma (HCC). Hepatocellular Carcinoma is among the most frequent malignant tumors of the liver. The risk of HCC development increases in parallel with the progression of liver fibrosis, which increases the need for HCC surveillance for patients with advanced fibrosis. In recent years, machine-learning techniques have been used as a prediction, classification, and diagnostic tools. They are easier, less time consuming, more accurate and effective approaches for early prediction of the risk of fibrosis and liver cancer development especially in the clinical decision support. This thesis aims to combine the serum biomarkers and clinical information to develop classification models for liver fibrosis stages. Particle swarm optimization, decision tree, multi-linear regression and genetic algorithm models for advanced fibrosis risk prediction were developed. These models predict the presence of advanced liver fibrosis with high accuracy and correlation coefficient, especially with the alternating decision tree and particle swarm optimization algorithms. The best model, using alternating decision tree alpha-fetoprotein (AFP) as predictor marker, achieved 86.2% negative predictive value (NPV), 0.78 area under the receiver operating characteristic curve (AUROC), and 84.8% accuracy on the test set, better than classical FIB-4 method. The uses of alpha-fetoprotein (AFP) as a feature of predicting advanced fibrosis in addition to using alternating decision tree improves the results than that of the FIB-4 algorithm which uses Alanine Aminotransferase (ALT) instead. In addition, a prediction model of the risk of HCC development for patients with CHC was established in this thesis. Alternating decision tree model achieved the highest AUROC 0.99. Classification and Regression Tree model achieved the highest positive predictive value, specificity, and accuracy values, which are 89.4%, 99.6%, and 98.7% respectively. The results lead to conclude that machine learning approaches could be used as a powerful, safe, and low-cost alternative for predicting strata of fibrosis and HCC developing rather than relatively risky alternative tools (such as the liver biopsy) in Chronic Egyptian Hepatitis C Virus Patients.
2018
Ph.d
Helwan
Engineering