A thesis submitted to the Faculty of Health Sciences, University of Witwatersrand, Johannesburg, in fulfilment of the requirements for the Degree of Doctor of Philosophy in the branch of Medicine: Diagnostic Radiology Johannesburg, 2017. / Introduction: Pneumonia is among the leading infectious causes of morbidity and mortality in children under 5 years globally. The chest X-ray remains the most readily available and most common imaging modality for the assessment of childhood pneumonia. Chest X-ray quality assurance is important to maintain high image quality, allowing for more accurate diagnosis. Chest X-ray patterns in a high HIV prevalence setting in children hospitalized with pneumonia in the era of bacterial conjugate vaccines has not been described. Furthermore the association between chest X-ray findings and microbiological etiology using novel models is important. Standardization of chest X-ray interpretation is important to allow comparison of research findings between studies and has been proposed by the World Health Organisation (WHO) as a measure of bacterial vaccine efficacy (VE). The role of Computer Aided Diagnosis (CAD) for chest X-ray interpretation is important, especially in countries with limited radiologists. Objectives: In this thesis we investigated the quality of chest X-rays and the impact of quality assurance activities in the Pneumonia Etiology Research for Child Health (PERCH) study, from 9 sites in 7 countries (Bangladesh, the Gambia, Kenya, Mali, South Africa, Thailand and Zambia), by quantifying radiographic errors through use of a customized quality assessment form. Within the context of the PERCH South African site, we compared chest X-ray patterns in HIV-unexposed, HIV- exposed-uninfected (HEU) and HIV-infected children hospitalized with WHO-defined severe or very severe community acquired pneumonia. Further, we evaluated the association of chest X-ray patterns with clinical parameters and its association with the PERCH Quantitative Analysis (PQA) model predicted probability of infection by a class or specific pathogens. Lastly, we evaluated the utility of CAD for identifying chest X-ray primary end-point pneumonia (CXR-PEP) versus non CXR-PEP compared to a consensus human interpretation as a reference standard. Results: A total of 747 chest X-rays in 9 PERCH sites had chest X-ray quality assessment. Collimation performed the worse of the 6 parameters, with 5 out of the 9 sites being graded sub-optimally for this. Three of the 9 sites showed a significant improvement in chest X-ray quality using a test for trend analysis. At the South African PERCH site, 920 cases were enrolled over two years, including 858 children with interpretable chest X-rays. The commonest finding was CXR-PEP, prevalent among 38% HIV-unexposed children, 33% HEU children and 60% of HIV-infected children, which was consistent between different age categories. CXR-PEP was twice as common in HIV-infected (OR 2.5; 95% CI 1.6-3.8) compared with these HIV-unexposed children. Clinical and laboratory features independently associated with CXR-PEP included the presence of severe malnutrition, fever and CRP > 40mg/dL. CXR-PEP was associated with the composite outcome of mechanical ventilation or death. However there was no single clinical or laboratory parameter that had both high sensitivity and specificity to discriminate between CXR-PEP from non CXR-PEP cases. HIV-unexposed children with a PQA model probability index > 0.5 for bacterial etiology had 2-fold greater odds (95% CI 1.1-4.1) for CXR-PEP compared to non CXR-PEP, and similarly HIV-unexposed children with microbiologically confirmed pneumococcal pneumonia (MCPP) had 3-fold greater odds (95% CI 1.3-6.7) of CXR-PEP. HIV-infected children with a PQA model probability index > 0.5 for respiratory viral etiology and PQA model probability index > 0.5 for Pneumocystis jiroveci etiology trended to have a greater odds for CXR-PEP vs non CXR-PEP. Using CAD4WHOKids for the 858 interpretable chest X-rays, for CXR-PEP versus non CXR-PEP, CAD4WHOKids had a sensitivity of 76%, specificity of 80% and area under the ROC curve of 0.850 (95% CI 0.823-0.876) compared to the radiologist consensus reading. Conclusion: Chest X-ray quality, in particular collimation was suboptimal in 7 resource limited countries, in children hospitalized for WHO defined severe or very severe pneumonia. CXR-PEP remains the most common chest X-ray abnormality in HIV-unexposed, HEU and HIV-infected children under 5 years of age hospitalized for WHO-defined severe or very severe pneumonia, even in the era of routine HiB and PCV immunization. Our findings support the literature that CXR-PEP is of limited use in HIV-infected children as an outcome measure in bacterial VE studies, due to opportunistic infections like Pneumocystis jiroveci pneumonia and respiratory viral infections contributing to CXR-PEP. CAD is promising as a diagnostic tool for identifying WHO defined CXR-PEP in bacterial VE trials and pneumonia epidemiological studies. / MT2017