C Young1, S Barker1, R Ehrlich2, B Kistnasamy3, A Yassi1. 1. School of Population and Public Health, University of British Columbia, Vancouver, BC, Canada. 2. Centre for Environmental and Occupational Health Research, School of Public Health and Family Medicine, University of Cape Town, Cape Town. 3. Department of Health, Pretoria, South Africa.
Abstract
BACKGROUND: For over one hundred years, the gold mining sector has been a considerable source of tuberculosis (TB) and silicosis disease burden across Southern Africa. Reading chest radiographs (CXRs) is an expert and time-intensive process necessary for the screening and diagnosis of lung disease and the provision of evidence for compensation claims. Our study explores the use of computer-aided detection (CAD) of TB and silicosis in CXRs of a population with a high incidence of both diseases. METHODS: A set of 330 CXRs with human expert-determined classifications of silicosis, TB, silcotuberculosis and normal were provided to four health technology companies. The ability of each of their respective CAD systems to predict disease was assessed using receiver operating characteristic curve analysis of the under the curve metric. RESULTS: Three of the four systems differentiated accurately between TB and normal images, while two differentiated accurately between silicosis and normal images. Inclusion of silicotuberculosis images reduced each system's ability to detect either disease. In differentiating between any abnormal from normal CXR, the most accurate system achieved both a sensitivity and specificity of 98.2%. CONCLUSION: The current ability of CAD to differentiate between TB and silicosis is limited, but its use as a mass screening tool for both diseases shows considerable promise.
BACKGROUND: For over one hundred years, the gold mining sector has been a considerable source of tuberculosis (TB) and silicosis disease burden across Southern Africa. Reading chest radiographs (CXRs) is an expert and time-intensive process necessary for the screening and diagnosis of lung disease and the provision of evidence for compensation claims. Our study explores the use of computer-aided detection (CAD) of TB and silicosis in CXRs of a population with a high incidence of both diseases. METHODS: A set of 330 CXRs with human expert-determined classifications of silicosis, TB, silcotuberculosis and normal were provided to four health technology companies. The ability of each of their respective CAD systems to predict disease was assessed using receiver operating characteristic curve analysis of the under the curve metric. RESULTS: Three of the four systems differentiated accurately between TB and normal images, while two differentiated accurately between silicosis and normal images. Inclusion of silicotuberculosis images reduced each system's ability to detect either disease. In differentiating between any abnormal from normal CXR, the most accurate system achieved both a sensitivity and specificity of 98.2%. CONCLUSION: The current ability of CAD to differentiate between TB and silicosis is limited, but its use as a mass screening tool for both diseases shows considerable promise.
Authors: Rodney Ehrlich; Stephen Barker; Jim Te Water Naude; David Rees; Barry Kistnasamy; Julian Naidoo; Annalee Yassi Journal: Int J Environ Res Public Health Date: 2022-09-29 Impact factor: 4.614
Authors: Jerry M Spiegel; Rodney Ehrlich; Annalee Yassi; Francisco Riera; James Wilkinson; Karen Lockhart; Stephen Barker; Barry Kistnasamy Journal: Ann Glob Health Date: 2021-07-01 Impact factor: 2.462