Development of diagnostic algorithm using machine learning for distinguishing between active tuberculosis and latent...

Area of Science:

• Medical Diagnostics
• Computational Biology
• Infectious Diseases

Background:

• Differentiating active tuberculosis (ATB) from latent tuberculosis infection (LTBI) is clinically challenging.
• Accurate diagnosis of Mycobacterium tuberculosis (Mtb) infection status is crucial for effective treatment and public health.

Purpose of the Study:

• To evaluate the diagnostic value of machine learning models using multiple laboratory data for distinguishing Mtb infection status.
• To develop and validate predictive models for differentiating ATB from LTBI.

Main Methods:

• Utilized T-SPOT, lymphocyte characteristic detection, and routine laboratory tests.
• Developed and compared 28 diagnostic models using various machine learning algorithms.
• Employed conditional random forests (cforest) as a primary modeling approach.

Main Results:

• Individual laboratory indicators showed limited diagnostic value (AUC < 0.8).
• 25 out of 28 machine learning models achieved AUC > 0.9 in the test set.
• The cforest model demonstrated superior performance with an AUC of 0.978 (sensitivity 93.39%, specificity 91.18%) in the discovery cohort, validated at 92.80% sensitivity and 89.86% specificity.

Conclusions:

• Machine learning, particularly the cforest model, offers a valuable and prospective tool for identifying Mtb infection status.
• This study presents a novel approach combining machine learning with laboratory findings for clinical diagnostic applications.