RT Dissertation/Thesis
T1 Computer-Aided Assessment of Tuberculosis with Radiological Imaging: From rule-based methods to Deep Learning
A1 Macías Gordaliza, Pedro
AB Tuberculosis (TB) is an infectious disease caused by Mycobacterium tuberculosis (Mtb.)that produces pulmonary damage due to its airborne nature. This fact facilitates the diseasefast-spreading, which, according to the World Health Organization (WHO), in 2021 caused1.2 million deaths and 9.9 million new cases.Traditionally, TB has been considered a binary disease (latent/active) due to the limitedspecificity of the traditional diagnostic tests. Such a simple model causes difficulties in thelongitudinal assessment of pulmonary affectation needed for the development of novel drugsand to control the spread of the disease.Fortunately, X-Ray Computed Tomography (CT) images enable capturing specific manifestationsof TB that are undetectable using regular diagnostic tests, which suffer fromlimited specificity. In conventional workflows, expert radiologists inspect the CT images.However, this procedure is unfeasible to process the thousands of volume images belongingto the different TB animal models and humans required for a suitable (pre-)clinical trial.To achieve suitable results, automatization of different image analysis processes is amust to quantify TB. It is also advisable to measure the uncertainty associated with thisprocess and model causal relationships between the specific mechanisms that characterizeeach animal model and its level of damage. Thus, in this thesis, we introduce a set of novelmethods based on the state of the art Artificial Intelligence (AI) and Computer Vision (CV).Initially, we present an algorithm to assess Pathological Lung Segmentation (PLS) employingan unsupervised rule-based model which was traditionally considered a neededstep before biomarker extraction. This procedure allows robust segmentation in a Mtb. infectionmodel (Dice Similarity Coefficient, DSC, 94%±4%, Hausdorff Distance, HD,8.64mm±7.36mm) of damaged lungs with lesions attached to the parenchyma and affectedby respiratory movement artefacts.Next, a Gaussian Mixture Model ruled by an Expectation-Maximization (EM) algorithmis employed to automatically quantify the burden of Mtb.using biomarkers extracted from thesegmented CT images. This approach achieves a strong correlation (R2 ≈ 0.8) between ourautomatic method and manual extraction. Consequently, Chapter 3 introduces a model to automate the identification of TB lesionsand the characterization of disease progression. To this aim, the method employs theStatistical Region Merging algorithm to detect lesions subsequently characterized by texturefeatures that feed a Random Forest (RF) estimator. The proposed procedure enables aselection of a simple but powerful model able to classify abnormal tissue.The latest works base their methodology on Deep Learning (DL). Chapter 4 extendsthe classification of TB lesions. Namely, we introduce a computational model to inferTB manifestations present in each lung lobe of CT scans by employing the associatedradiologist reports as ground truth. We do so instead of using the classical manually delimitedsegmentation masks. The model adjusts the three-dimensional architecture, V-Net, to a multitaskclassification context in which loss function is weighted by homoscedastic uncertainty.Besides, the method employs Self-Normalizing Neural Networks (SNNs) for regularization.Our results are promising with a Root Mean Square Error of 1.14 in the number of nodulesand F1-scores above 0.85 for the most prevalent TB lesions (i.e., conglomerations, cavitations,consolidations, trees in bud) when considering the whole lung.In Chapter 5, we present a DL model capable of extracting disentangled information fromimages of different animal models, as well as information of the mechanisms that generatethe CT volumes. The method provides the segmentation mask of axial slices from threeanimal models of different species employing a single trained architecture. It also infers thelevel of TB damage and generates counterfactual images. So, with this methodology, weoffer an alternative to promote generalization and explainable AI models.To sum up, the thesis presents a collection of valuable tools to automate the quantificationof pathological lungs and moreover extend the methodology to provide more explainableresults which are vital for drug development purposes. Chapter 6 elaborates on theseconclusions.
YR 2022
FD 2022-02
LK https://hdl.handle.net/10016/35380
UL https://hdl.handle.net/10016/35380
LA eng
NO Mención Internacional en el título de doctor
DS e-Archivo
RD 27 jul. 2024