Two-centers machine learning analysis for predicting acid-fast bacilli results in tuberculosis sputum tests.

Area of Science:

• Medical diagnostics
• Infectious disease research
• Machine learning in healthcare

Background:

• Tuberculosis (TB) is a chronic respiratory infection caused by Mycobacterium tuberculosis.
• Current diagnosis relies on sputum smear microscopy for acid-fast bacilli (AFB) to assess infectivity.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting AFB results in TB patients.
• To identify key factors from blood samples for TB diagnosis using ML.

Main Methods:

• Enrolled 769 patients (641 training, 128 validation).
• Analyzed blood samples using various ML methods.
• Selected Support Vector Machine Recursive Feature Elimination (SVM-RFE) to build a nomogram diagnostic model.

Main Results:

• The ML diagnostic model achieved an Area Under the Curve (AUC) of 0.721 (training) and 0.758 (validation).
• Decision curve analysis indicated clinical utility within a specific threshold range.

Conclusions:

• A novel ML-based diagnostic model was developed for predicting AFB results in TB.
• The model demonstrated satisfactory diagnostic performance and clinical utility.