Table 1.
Classification of study population according to clinical and microbiological data.
Figure 1.
Flow chart of individuals taking part in the study.
Table 2.
Summary statistics of study population.
Figure 2.
Distribution of CAD scores for patient groups A (s+/c+ M.tb), B (s−/c+ M.tb), C (s±/c+ NTM), D (s−/c− clin.TB) and F (Controls).
Table 3.
Median CAD scores and 90% central range.
Figure 3.
ROC analysis for the detection of M.tb culture-positive individuals.
Legend. —–A (s+/c+ M.tb), B (s−/c+ M.tb) vs. F (Controls): Az = 0.84 (0.80–0.88), - - - A (s+/c+ M.tb), B (s−/c+ M.tb) vs. C (s±/c+ NTM), E (EPTB), F (Controls): Az = 0.81 (0.77–0.85), p = 0.28.
Figure 4.
ROC analysis for the detection of M.tb culture-positive individuals by smear status.
Legend. —– A (s+/c+ M.tb) vs. F (Controls): Az = 0.90 (0.86–0.93), - - - B (s−/c+ M.tb) vs. F (Controls): Az = 0.67 (0.58–0.75), p<0.01.
Figure 5.
ROC analysis for the detection of M.tb culture-positive individuals by HIV Status.
Legend. —– HIV negative. A (s+/c+ M.tb), B (s−/c+ M.tb) vs. F (Controls): Az = 0.89 (0.85–0.94), - - - HIV positive. A (s+/c+ M.tb), B (s−/c+ M.tb) vs. F (Controls): Az = 0.79 (0.72–0.86), p<0.01.
Table 4.
Performance of CAD4TB and human readers.
Figure 6.
Comparison of automated and human reading.
Legend. Sensitivity and specificity to distinguish group A (s+/c+ M.tb) and B (s−/c+ M.tb) vs. F (Controls). Line and shaded area: ROC curve and 95% CI for CAD4TB. The expert reader is represented by square symbols, the clinical officer by round symbols. The different fill of the symbols indicate different reading thresholds: empty symbols = ‘any abnormality’, crossed symbols = ‘TB consistent abnormalities’ and filled symbols = ‘abnormalities highly suggestive for TB’.