A Deep Multi-Model Approach for Automatic Tuberculosis Severity Assessment with Computed Tomography Images

The thesis addressed the problem of tuberculosis analysis in lung CT image scans. It proposed methodologies for segmenting tuberculosis-affected areas of the lung and the severity assessment of the disease. Three models were proposed, (i) the 3D DAvoU-Net model for segmentation (ii) The 3D-CNN Feature Learning Model and (iii), The Fusion Model. All three models were evaluated on ImageCLEF 2019 and 2021 tuberculosis datasets. the models were further tested for generality using the COVID-19 dataset. All three models produced outstanding results in segmenting and classifying TB and COVID-19 lung CT slices.