AI accelerates TB drug discovery and treatment monitoring

In a transformative leap for global health, artificial intelligence (AI) is reshaping the fight against tuberculosis (TB), an infectious disease that continues to claim over a million lives annually. A recent review study titled “Integration of AI and ML in Tuberculosis (TB) Management: From Diagnosis to Drug Discovery” published in the journal Diseases (2025) outlines how AI and machine learning (ML) technologies are revolutionizing nearly every phase of TB management - from early detection to treatment adherence and the discovery of novel drugs.

How are traditional TB diagnostics falling short?

Despite longstanding reliance on sputum smear microscopy, culture methods, and chest X-rays, conventional TB diagnostics remain plagued by delays, inaccuracies, and resource limitations. These methods often require weeks to yield results and demand expensive infrastructure such as biosafety level 3 laboratories. Sensitivities vary widely, with sputum smear microscopy ranging from 42–63% and mycobacterial cultures taking up to six weeks to confirm infection. Additionally, these tools are less effective in identifying extrapulmonary TB or cases with low bacterial loads.

The shortcomings of these traditional techniques are exacerbated in under-resourced and high-burden regions, where the lack of trained radiologists and specialized equipment limits diagnostic reach. Moreover, tools like nucleic acid amplification tests (NAATs) and the GeneXpert system, while more accurate, struggle to distinguish viable from non-viable bacilli and require significant investment for widespread deployment.

AI, with its capacity to automate, scale, and optimize diagnostic procedures, offers a compelling solution to these limitations. Through deep learning models such as convolutional neural networks (CNNs), AI can interpret medical imaging with a level of precision that rivals expert radiologists, enabling early and accessible detection across diverse healthcare settings.

How is AI reshaping TB diagnosis and prognosis?

Artificial intelligence is not just improving diagnostic accuracy - it is redefining the diagnostic paradigm. CNNs and other deep learning architectures are being trained on vast datasets of annotated chest X-rays, enabling rapid, high-sensitivity identification of TB-related abnormalities such as infiltrates and cavities. In many instances, AI-assisted tools have achieved diagnostic accuracy exceeding 90%, significantly outperforming traditional radiographic evaluations.

In countries such as Vietnam, India, and Myanmar, AI-driven X-ray interpretation systems have been deployed for community-wide TB screening with notable success. In India, Wadhwani AI's technologies have enhanced screening efficiency and improved outcome predictions, while Myanmar’s integration of AI tools has compensated for the acute shortage of radiologists.

Beyond imaging, AI is playing a crucial role in molecular diagnostics. Machine learning models are being integrated into CRISPR–Cas systems and GeneXpert assays to interpret genetic data and identify resistance markers. These models facilitate faster and more accurate identification of drug-resistant strains, critical in curbing the spread of multidrug-resistant TB (MDR-TB).

Moreover, AI-assisted diagnosis from computed tomography (CT) scans is pushing the frontier even further. Using 3D-CNNs, AI models can interpret volumetric datasets to quantify lesion characteristics and monitor disease progression with high sensitivity - capabilities that were previously limited to specialized centers.

Can AI improve TB treatment monitoring and drug discovery?

AI’s influence extends well beyond detection. In treatment monitoring, AI is streamlining patient adherence tracking and side effect surveillance, both vital for curbing treatment failure and resistance development. Digital Adherence Technologies (DATs) powered by AI, such as video-based directly observed therapy (VDOT), are allowing healthcare providers to remotely verify medication intake, particularly in underserved or rural areas. These systems enhance patient autonomy, reduce costs, and improve treatment compliance.

AI is also proving invaluable in predicting treatment outcomes. Studies have demonstrated that AI models, such as random forests and support vector machines, can forecast the efficacy of treatment and the likelihood of adverse events with accuracies surpassing 80%. For example, the CNN model has shown the lowest mean absolute error (MAE) in estimating optimal treatment durations, a vital consideration in customizing therapy for each patient.

In the realm of drug discovery, AI is accelerating the development of new anti-TB therapies by identifying novel drug targets and optimizing compound screening. Machine learning algorithms are being used to model protein-ligand interactions, predict toxicity, and guide compound repurposing. AI has enabled virtual screening of FDA-approved drugs against Mycobacterium tuberculosis, uncovering promising candidates like bedaquiline and delamanid that are now essential components of MDR-TB regimens.

Drug repurposing strategies using AI have also gained traction, offering a cost-effective and faster path to identifying therapeutics for TB. AI models can detect hidden pharmacological relationships by mining chemical databases and historical screening results, dramatically narrowing down viable drug candidates.

What are the challenges and what lies ahead?

Despite its promise, AI integration in TB management is not without hurdles. Data quality and diversity remain major concerns, especially in developing standardized, bias-free datasets for training robust models. Infrastructure limitations, ethical considerations surrounding data privacy, and the interpretability of AI decisions present additional barriers to widespread adoption.

Furthermore, many AI models lack real-world validation across diverse patient populations. Without rigorous field trials and context-specific calibration, their utility remains constrained. AI is also limited in its ability to consider clinical nuances such as symptom history and comorbidities, which can affect diagnostic accuracy and lead to false positives.

The path forward requires comprehensive policy frameworks, stakeholder collaboration, and sustained investment in research and infrastructure. Regulatory bodies must establish clear guidelines for the ethical deployment of AI tools in clinical environments. Simultaneously, future research must focus on refining algorithmic precision, enhancing explainability, and aligning AI applications with national TB control strategies.