Tuberculosis (TB) kills approximately 1.6 million people yearly. Among the three top infectious killers, TB is rising worldwide while HIV-AIDS and malaria are trending down, despite the fact anti-TB drugs
are generally curative. The problem lies in the inefficient detection of this complex disease. It is
hypothesized utilizing informatics methods such as deep learning (DL) approaches to machine learning
(ML) analysis of radiological data, combined with clinical, microbiological, and immunological data,
delivered as clinical decision support (CDS), can not only afford better diagnostics but also improved
determination of pathogenesis and severity, and monitoring efficacy of therapy. This study proposes a
comprehensive approach for efficient disease detection employing informatics methods including but not
limited to information retrieval, visualization, ML and artificial intelligence (AI), tested in nonhuman
primate (NHP) model which captures human TB most closely, as a prelude to TB patient studies. A group
of six rhesus macaques were experimentally inoculated with Mycobacterium tuberculosis (M. tb., Erdman
strain) in the right lower lung. Animals were followed at regular intervals over 24 weeks by Computed
Tomography (CT) imaging. A DL algorithm was developed and trained for automated scoring of lung CT
images and compared head-to-head with radiologists’ scores. Correlations of ML scores with several
other TB indicators were also performed. DL model afforded early disease detection as compared to
radiologists. Importantly, ML analysis demonstrated greater consistency over multiple runs compared to
scoring by two radiologists. ML scores also exhibited strong correlations with granuloma and total TBlesion volumes at necropsy, and disease-signs and blood biomarkers throughout pathogenesis. ML-based
analysis of radiological imaging enabled early and consistent disease detection and assessment of
severity, enabling a noninvasive and objective approach. ML and AI approaches can improve early
detection and understanding of the disease. In addition, the multimodality approach described here is
valuable in monitoring efficacy of therapy.