Although tuberculosis (TB) is relatively rare in Canada, this lung disease continues to disproportionately affect certain populations, particularly Inuit and other Indigenous Peoples. It also remains common among people born outside of Canada who, before immigrating, were exposed to TB in countries with higher infection rates. The incidence of TB is strongly influenced by social determinants of health, such as poverty, inadequate housing and barriers to accessing health care. The disease is curable and preventable but can be fatal if left untreated.
Through the Government of Canada's tuberculosis response, our country has committed to eliminating TB by improving early detection, access to care and treatment. Early detection—usually with a simple skin prick, blood test or chest X-ray—is key for treating TB. However, in remote or underserved regions where TB rates are high, early detection is difficult because of limited resources and barriers to accessing health care.
To help address this challenge, Dr. Ashkan Ebadi, a researcher with the NRC's Digital Technologies Research Centre, partnered with Dr. Alexander Wong, a University of Waterloo professor and co-director of the Vision and Image Processing Research Group. Together, they developed an algorithm that can analyze chest X-rays to screen and detect cases of TB. This technology could transform how health-care providers in remote areas screen for the disease, enabling faster and more accurate diagnoses.
A unique feature of the AI tool is its ability to learn on its own, with minimal human supervision. Using a teacher–student set-up, an initial "teacher" model is trained on a small set of X-rays, including confirmed cases of TB. Once the teacher model becomes good enough, it begins training a "student" model on a larger set of unknown images. Just like when humans learn, the teacher model provides constant feedback as the student model scans new images looking for cases of TB. Early results reported in a 2024 IEEE paper by Patel, Wong and Ebadi showed that their AI model diagnosed TB with a 98% accuracy rate.
The model also uses logic and reasoning behind its decision making—a unique feature called explainable AI, or explainability. Unlike more traditional "black box" models, an explainable AI model describes the reasoning behind an answer or, in this case, a diagnosis. When analyzing an X-ray, the model highlights regions that are relevant for a TB diagnosis. Explainability is crucial in high-stakes areas such as health care. It allows health-care workers to build confidence in the system and provides transparency behind the AI model's automated decision making.
This project builds on earlier successes. During the COVID-19 pandemic, Dr. Ebadi and his collaborators developed COVID-Net, AI-based solutions for rapid diagnosis of COVID-19 and determining the prognosis of infected patients. The models gained international recognition and were used by researchers and hospitals around the world. With COVID-19 cases now under better control, the team saw an opportunity to apply their AI technology to other health-care challenges, especially ones where there is an immediate need.
Looking ahead, Dr. Ebadi hopes this AI model can help overcome barriers to TB screening, and early detection in remote and underserved regions in Canada by reducing the need for access to specialized staff. Although their solution shows early promise, the team now needs to partner with collaborators in hospital settings and use real-world population data. This step is essential for fine-tuning and ensuring accuracy in a clinical setting.
This AI-based technology developed by Dr. Ebadi and his collaborators has the potential to bring advanced diagnostic tools to underserved communities where early detection of TB can make a big difference.