[R21] Integrated computer-aided, point-of-care ultrasound for tuberculosis screening

点击翻译按钮获取中文摘要

PROJECT ABSTRACT Mycobacterium tuberculosis (TB, Mtb) is one of the leading causes of infectious disease mortality worldwide, with a persistent gap in global case detection. In 2020, only 5.8 million of the estimated 9.9 million individuals who became ill with TB were diagnosed and reported. Active case finding efforts to find these undetected cases by screening individuals in the community for TB is not feasible in most resource-limited settings due to the large number needed to screen with sputum microbiologic tests to detect one case. A rapid, highly sensitive point-of-care test that can be performed in the field to screen for high-risk individuals can improve access to active case finding by substantially reducing the number needed to test. Despite progress in the field, there remains a lack of such tests that achieve both the point-of-care characteristics accessible for large-scale screening and the target accuracy profiles for a triage or rapid diagnostic test. Point-of-care ultrasound (POCUS) devices are low-cost, portable, avoid radiation exposure, and do not require trained radiologic staff, making them amenable to wide scale with minimal resource needs. Lung ultrasound has been found to diagnose adult pneumonia at a comparable to improved performance level than chest-X-ray and holds promise as a TB triage test. Further application of artificial intelligence-based computer aided diagnosis algorithms can further improve reproducibility by automating and standardizing image interpretation, and potentially improve performance as evolving AI platforms train on growing datasets. To explore the utility of POCUS CAD for TB screening, we will conduct a systematic evaluation of POCUS for TB screening among TB household contacts across a spectrum of early to advanced infection phenotypes. We will additionally pilot an integrated computer-aided detection algorithm for POCUS-based detection of pulmonary TB with the integration of clinical, exposure, and prevalence data to explore whether these variables can increase specificity for TB. We will then evaluate performance profiles using both deconstructed feature-based analysis and deep-learning algorithms to inform the potential utility of POCUS as a TB triage test to enable widespread active case finding for the millions of undiagnosed TB cases.