Using Particle Swarm Optimization to Find Efficient Designs for Mixed Effects Models with Sparse Grid and Predict Progression of Idiopathic Pulmonary Fibrosis using Baseline High Resolution Computed Tomography Scans with Random Forest
There are many challenging optimization problems in the health sciences. Problems in health sciences are increasingly complex, and frequently the most advanced optimization techniques are required to tackle them. Researchers thus need various types of flexible optimization tools that are easily accessible and efficient. In this dissertation, we utilize a stochastic optimization technique called particle swarm optimization (PSO) and demonstrate its usefulness and flexibility using two applications in the biomedical field. For the first application, we propose a sparse grid hybridized PSO (SGPSO) algorithm to find different types of optimal or highly efficient designs for various longitudinal models. In particular, we consider non-linear mixed effects models useful in pharmacokinetic/pharmacodynamic studies and show SGPSO is a powerful tool for finding optimal or efficient designs that were previously thought to be intractable. For the second application, we propose a random forest hybridized quantum PSO algorithm for predicting disease progression of idiopathic pulmonary fibrosis (IPF) using quantitative information on high-resolution computed tomography (HRCT) imaging. IPF is a fatal type of lung disease with unpredictable functional progression at the time of diagnosis. We leverage single time point HRCT scans to predict the 6 months to 1 year follow-up status of IPF subjects. Results show that the two hybridized PSO approaches tackle important biomedical optimization problems effectively, and since the proposed methodology is not problem specific, there is potential for further applications to solve other biomedical optimization problems.