While there has been progress in controlling many diseases in the past decade in the United States, chronic liver disease mortality rate has increased according to Center of Disease Control and Prevention. There is an urgent need for developing new technologies and more sensitive methods for early diagnosis of liver disease. This thesis proposes quantitative magnetic resonance imaging techniques to leverage the existing methods and increase sensitivity and specificity for liver disease diagnosis. We proposed a numerical optimization technique for efficient and more accurate calculation of perfusion maps in the liver. We proposed a new algorithm for water/fat separation resulting in a rapid robust method of quantitative susceptibility mapping (QSM) in the liver. We investigated deep learning approaches in water/fat separation to replace the current classical solvers and finally we investigated a machine learning approach to use QSM and R2* to enhance fibrosis detection. The proposed frameworks were tested in both healthy volunteers and patients for cancer diagnosis, iron measurement and fibrosis detection with the possibility of translation into clinical practice.