Machine learning approaches in diagnosing tuberculosis through biomarkers - A systematic review

Area of Science:

• Biomedical informatics
• Computational biology
• Infectious disease diagnostics

Background:

• Tuberculosis (TB) diagnosis can be accelerated by biomarker-based tests, improving patient outcomes.
• Machine learning (ML) offers a promising avenue for enhancing TB detection accuracy and speed.

Approach:

• A systematic review following PRISMA guidelines identified 19 studies on ML for biomarker-based TB diagnosis.
• Supervised learning, particularly Support Vector Machine (SVM) and Random Forest algorithms, dominated the reviewed literature.

Key Points:

• Highest reported accuracy, sensitivity, and specificity for ML models were 97.0%, 99.2%, and 98.0%, respectively.
• Protein and gene-based biomarkers (RNA sequence, Spoligotypes) were commonly investigated.
• Leave-one-out cross-validation was frequently used to prevent model overfitting.

Conclusions:

• ML demonstrates significant potential to improve TB diagnosis via biomarkers, offering a faster alternative to conventional methods.
• Biomarker-based ML models could be particularly impactful in low- and middle-income countries where sputum-based tests are less accessible.