AI-based System for Lung Tuberculosis Screening: Diagnostic Accuracy Evaluation

AI-based System for Lung Tuberculosis Screening: Diagnostic Accuracy Evaluation

Testing of AI solutions to assess diagnostic accuracy for tuberculosis detection.

Tuberculosis remains a key problem of modern medicine. New approaches for burden overcoming should be proposed. New screening strategies may include artificial intelligence (AI). An AI-based system for chest x-ray analysis and triage ("normal/tuberculosis suspected") have been developed and trained. A special data-set was prepared. There are 238 normal x-rays and 70 x-rays with lung tuberculosis in data-set. The data-set was randomly divided into 2 samples:

• sample N1 (n=140) with ratio "normal: tuberculosis" 50:50,
• sample N1 (n=150) with ratio "normal: tuberculosis" 95:5. Both samples will be analysed by AI-based system. Results will be quantified using diagnostic accuracy metrics: sensitivity and specificity, positive and negative predictor values, likelihood ratio, and area under the ROC (receiver operating characteristic) curve.

Inclusion Criteria:

• no pathology in a lung on chest x-ray
• signs of lung tuberculosis on chest x-ray

Exclusion Criteria:

• any pathology in the lungs (except tuberculosis)