Tuberculosis (TB) screening may soon benefit from a new ally: artificial intelligence (AI). Recent research suggests that AI-powered tools can detect signs of active pulmonary tuberculosis on chest X-rays with accuracy comparable to that of experienced radiologists, offering a potential solution to diagnostic challenges in resource-limited settings.
The Challenge of TB Diagnosis
TB remains a significant global health concern, particularly in regions with limited access to specialized medical expertise. Accurate and timely diagnosis is crucial for effective treatment and preventing further spread of the disease. However, interpreting chest X-rays, a common diagnostic tool for TB, requires skilled radiologists, who are often in short supply in high-burden areas.
New Research Validates AI’s Potential
A retrospective analysis of nearly 500 chest X-ray films collected in Guinea-Bissau and Ethiopia between and investigated the performance of computer-aided detection (CAD) chest X-ray software. The images, some captured using mobile phones and digital cameras, were assessed both by the CAD software and independently by two Ethiopian radiologists. A definitive TB diagnosis was based on clinical and laboratory findings, including detection of Mycobacterium tuberculosis using the Xpert MTB/RIF assay.
The study found that the AI software demonstrated an area under the receiver operating characteristic curve (AUC) of 0.84 for identifying Xpert-confirmed pulmonary tuberculosis. At a predefined cut-off point, the software achieved a sensitivity of 76.5% and a specificity of 85.9%. For comparison, Radiologist A had a sensitivity of 64.7% and specificity of 91.9%, while Radiologist B showed a sensitivity of 76.5% and specificity of 82.3%.
Agreement Between AI and Human Readers
The level of agreement between the AI software and the radiologists was considered moderate. Agreement between the two radiologists themselves was κ=0.45, while agreement between each radiologist and the AI software was κ=0.56. This suggests a reasonable level of consistency between the AI’s interpretations and those of trained medical professionals.
Implications for Resource-Constrained Settings
These findings are particularly encouraging for regions facing shortages of radiological expertise. The ability of AI to accurately interpret chest X-rays, even from images captured with readily available devices like mobile phones, could significantly improve TB screening programs. The research suggests that AI could serve as a valuable triage tool, helping to prioritize cases for review by radiologists and ensuring that patients receive timely and appropriate care.
Further research, published in , validated the use of AI for detecting active pulmonary TB and other chest X-ray abnormalities in a multi-site prospective study. Researchers found that AI matched the performance of radiologists in detecting active tuberculosis and also showed promise in identifying other chest X-ray abnormalities effectively. This study, conducted in settings with high TB and HIV prevalence, aimed to evaluate a commercially available chest X-ray-based AI system in real-world clinical settings.
A separate study, evaluating an AI model called AIRIS-TB, demonstrated even more impressive results. AIRIS-TB was evaluated on over one million chest X-rays, achieving an AUC of 98.51% and an overall false negative rate of 1.57%, outperforming radiologists (1.85%) while maintaining a 0% TB-false negative rate. The model has the potential to automate up to 80% of routine chest X-ray reporting by selectively deferring only cases with findings to radiologists.
How AI Aids in TB Detection
The integration of AI into radiology workflows can free up medical staff to focus on more complex cases, particularly in settings with limited resources. By improving the interpretation process, AI-assisted solutions have the potential to revolutionize TB detection and improve patient outcomes. AI can also assist in identifying other chest X-ray abnormalities, broadening its potential impact on respiratory health.
The Role of Radiologists Remains Crucial
While AI shows great promise, it is not intended to replace radiologists entirely. The technology is best viewed as a tool to augment their expertise, helping them to work more efficiently, and accurately. Radiologists will continue to play a vital role in confirming diagnoses and managing complex cases. The study emphasizes that AI can be particularly helpful in triaging cases, allowing radiologists to focus their attention on those most in need of their expertise.
In cases where a radiologist identifies signs suggestive of TB on a chest X-ray, this is referred to as ‘suspected TB-positive’, regardless of whether sputum-based confirmatory lab tests are positive.
Looking Ahead
The development and validation of AI-powered tools for TB detection represent a significant step forward in the fight against this global health challenge. As AI technology continues to evolve, it is likely to play an increasingly important role in improving access to accurate and timely TB diagnosis, particularly in resource-constrained settings. Further research will be crucial to refine these tools and ensure their effective implementation in diverse healthcare systems.
