Computational tools for the radiological characterization of tuberculosis in X-Ray computer tomography images.

Tuberculosis (TB) remains the world’s second-deadliest disease, after HIV/AIDS, due to a single infectious agent. However, its drug development pathway is not completely integrated having important unanswered questions at every stage. This Bachelor Thesis is framed in an international project that aims to overcome the gap in TB drug development. For that purpose, they are pioneering a novel integrated methodology for preclinical trials, capable of facilitating the transition of the best combination of drugs to clinical trials and maximizing their chances of success.

In this context, the developed work aims to provide an array of innovative enhancements to their model systems. Its main objective is twofold: 1) To evaluate the performance of a tool, dedicated to the semi- automatic segmentation of TB infected lungs in Computed Tomography images; 2) to extract a TB biomarker based on the intensity differences between healthy and sick lung parenchyma from CT images, capable of characterizing disease longitudinal evolution. The performance evaluation was carried out by an inter-observer analysis. Our goal was to confirm that independently of User characteristics, segmentations would turn out to be similar between them, while at the same time being precise. The extraction of the TB biomarker was performed by manual thresholding. However, this procedure was later on automated by means of the Estimation Maximization (EM) algorithm for Gaussian mixtures, in order to obtain more objective and robust results.

Regarding inter-observer analysis, results showed that the tool had a better performance in subjects with low degree of infection. Furthermore, regarding the extraction of the TB biomarker, results showed that some of the extracted volumes were capable of characterizing disease longitudinal evolution. Nevertheless, being our sample size quite small, further studies are necessary in order to reinforce our assumptions.