%0 PDF %T Mathematical and Machine Learning Approaches to Predicting Drug Penetration in Heterogeneous Tuberculosis Lesions. %A Rayfield, Adam. %D 2018-05-16 12:50:01 -0400 %8 2018-05-16 %I Tufts Archival Research Center %R http://localhost/files/8049gh348 %X Tuberculosis is among the most widespread infectious diseases in the modern world. The disease is characterized by the lesions, or granulomata, which its infection form in the lungs, which are resilient to antibiotic penetration and can cause latent, chronic infections. Current research aims to improve predictions of tuberculosis disease outcomes and improve therapy by studying tuberculosis through animal models, in humans, and in computational simulations of mathematical models. The abundance of drug distribution image data available from animal and human sources is a target for machine learning techniques, which could assist in predicting the outcomes of disease treatments on specific lesions, and prior models may inform the design of new mathematical models which incorporate spatially-relevant information, a necessity for predictions involving infected granulomata. The prospects of convolutional neural networks, a k-nearest neighbor algorithm, and a mathematical model in COMSOL Multiphysics for generating predictions relevant to clinical outcomes are examined, and these examined methods show promise to be developed further in the future. %G eng %[ 2022-10-07 %~ Tufts Digital Library %W Institution