Contribution of artificial intelligence to the imaging diagnosis of pediatric pulmonary tuberculosis

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Infectious Diseases

Background:

• Pediatric tuberculosis (TB) diagnosis presents significant challenges globally and in Brazil.
• The Brazilian Ministry of Health utilizes a clinical scoring system (S-MoH) for suspected pediatric TB, often requiring specialist interpretation of radiographs.
• AI systems like CAD4TB are WHO-approved for adult TB screening but lack recommendations for children under 15.

Purpose of the Study:

• To evaluate the performance of CAD4TBv7.1 as a screening tool for pediatric TB in Brazil.
• To compare CAD4TBv7.1 performance against microbiological confirmation and the S-MoH scoring system.
• To assess the potential of AI as a complementary diagnostic aid in pediatric TB.

Main Methods:

• A retrospective study analyzed 179 chest radiographs from patients aged 0-14 years with suspected pulmonary TB.
• CAD4TBv7.1 was used to analyze radiographs, with two Youden's index-derived cutoff points applied for comparison.
• Results were benchmarked against microbiological confirmation and the S-MoH score.

Main Results:

• CAD4TBv7.1 achieved an Area Under the ROC Curve (AUROC) of 0.71 against microbiological diagnosis (sensitivity 52%, specificity 86.3%).
• Against the S-MoH score, CAD4TBv7.1 showed an AUROC of 0.59 (sensitivity 34.43%, specificity 86.44%).
• The AI demonstrated low sensitivity but high specificity, indicating potential as a screening tool.

Conclusions:

• CAD4TBv7.1 shows potential as a complementary screening tool for pediatric TB, particularly in resource-limited settings.
• Standalone use of CAD4TBv7.1 in children under 15 is not yet recommended.
• Further pediatric validation and integration with clinical approaches are necessary for effective AI implementation in pediatric TB diagnosis.