Objective:

To develop and validate a multimodal machine-learning framework for differentiating Mycobacterium tuberculosis lung disease (MTB-LD) from nontuberculous mycobacterial lung disease (NTM-LD) using clinical symptoms, hematological biomarkers, and high-resolution computed tomography (HRCT) features.

Approach:

• Study Design: A retrospective study involving 102 patients with microbiologically confirmed mycobacterial lung disease, including 53 with MTB-LD and 49 with NTM-LD.
• Machine Learning Framework: An interpretable multimodal machine-learning framework was developed, integrating clinical symptoms, laboratory biomarkers, and HRCT features, evaluated using k-nearest neighbors, logistic regression, and random forest classifiers.

Key Findings:

• Multimodal integration of HRCT, clinical, and laboratory features outperformed single-modality approaches.
• The random forest model achieved the best performance with an AUC of 0.92, sensitivity of 0.89, specificity of 0.93, and F1-score of 0.90.
• Key predictive contributors included cystic bronchiectasis, tree-in-bud sign, fever, and selected laboratory biomarkers.

Interpretation:

The findings indicate that multimodal clinical data may assist in differentiating MTB-LD from NTM-LD, but the framework requires external validation before clinical use.

Limitations:

• The study is retrospective and conducted at a single institution, which may limit generalizability.
• External validation of the framework is necessary before clinical implementation.

Conclusion:

The study presents a multimodal machine-learning approach for differentiating mycobacterial lung diseases, highlighting the need for further validation.