In Russia, AI Will Detect Tuberculosis on X-Ray Images with Near-Perfect Accuracy

Poor quality is not an obstacle

A research team at Lomonosov Moscow State University has introduced a diagnostic approach that strengthens AI accuracy even when X-ray images are noisy or low resolution. The method builds on Fast and Adaptive Bidimensional Empirical Mode Decomposition (FABEMD), a technique for data augmentation that removes background artifacts while preserving clinically meaningful features. This allows AI models to train on a wider range of image variations and develop greater resilience to image quality issues.

When tested on major international datasets, including Montgomery, Shenzhen, and TBX11K, the inclusion of synthetic images boosted classification accuracy—especially on challenging cases where datasets were limited. According to the researchers, the results show that AI systems are beginning to achieve a diagnostic performance comparable to human interpretation.

Helping doctors

The new method is particularly valuable for regions where experienced radiologists are scarce or available hardware produces low-quality scans. In these settings, AI can serve as a reliable first-line screening tool, quickly highlighting suspicious cases for further review.

The authors emphasize that AI should not replace physicians. Instead, it is designed as an assistive technology that accelerates screening and reduces diagnostic burden while keeping final medical decisions in human hands.

Not just tuberculosis

The FABEMD-based approach to data enhancement may support diagnostics beyond tuberculosis. Researchers believe it could be applied to lung disease detection, ophthalmic diagnostics, oncology, and other clinical areas that rely on imaging.

The technology also supports broader adoption of AI in healthcare by helping systems process large volumes of scans and freeing clinicians to focus on treatment and complex diagnostic tasks.