The use of computer-aided detection for tuberculosis (TB) diagnosis in the Bono East, Ghana

Background: Tuberculosis (TB) continues to pose a significant public health challenge annually.In the World Health Organization African and Americas regions, 81% of TB cases were diagnosed bacteriologically through sputum smear microscopy or culture.While bacteriological confirmation is ideal due to its ability to detect drug resistance and inform effective treatment, limited resources and diagnostic capacity have constrained its widespread use.In the Bono East region of Ghana, where diagnostic coverage is limited, the regional TB case detection rate was only 44% in 2023.This study aimed to enhance TB case detection rates and assess the effectiveness of a computer-aided detection CAD4TB technology in diagnosing TB by comparing its results with radiologist-confirmed diagnoses.Methods: Four districts with the lowest TB detection rates and limited access to GeneXpert machines were targeted.A total of 250 presumed TB clients were screened using symptom-based assessment and digital chest X-ray imaging.The images were analyzed using CAD4TB Version 7, an artificial intelligence-based tool that assigns scores from 0 to 100 indicating the likelihood of TB-related lung abnormalities.All images were reviewed by a senior radiologist and classified as normal or suggestive Results: Of the 250 individuals screened, CAD4TB identified 23 (9.2%) possible TB cases.The majority, 17 (73.9%),were classified as having a "Very High" likelihood of TB, with a score ranging from 71-100.Comparison with radiologist interpretations showed strong agreement for normal cases (223 cases; mean score: 17.16) and suggestive of Pulmonary TB (17 cases; mean score: 81.61).However, discrepancies were observed, including four false negatives (mean score: 42.50) and six false positives (mean score: 67.71), highlighting limitations in diagnostic precision.Conclusion: These findings demonstrate that CAD4TB is a promising supplementary tool for TB detection in resource-limited settings, supporting early identification of cases that may otherwise be missed in the absence of GeneXpert testing.