Leveraging Deep Learning for Automated Screening of Pulmonary Tuberculosis using Genki

Introduction : Tuberculosis claims more lives than any other infectious disease and ranks amongst the top 10 causes of death globally. The End TB Strategy proposed by WHO puts a renewed emphasis on the early detection of TB. Due to the high costs associated with molecular tests for diagnosing TB, using digital chest X-rays to screen cases and subsequently subjecting them to molecular or sputum tests is an affordable course of action more accessible to priority groups. In recent years, computer-aided detection (CAD) systems have taken advantage of three technology trends: rapid developments in the field of deep learning, availability of large annotated data to train predictive models, and easy access to high computing power owing to cloud computing and GPUs.

The Genki system developed at DeepTek leverages these trends and the latest research in artificial intelligence and machine learning to automatically screen digital chest X-rays for signs of TB.

Results : 

• In our study, a group of radiologists assisted by Genki demonstrated 86% improvement in productivity and 46% reduction in reporting turnaround time over the control group not using Genki.

• Genki detects TB in a patient with an accuracy of 92%, sensitivity of 91% and specificity of 92% with an AUROC of 96%.

Conclusion : We believe that the human-in-the-loop approach that uses AI to pre-

screen X-rays but leaving the final decision to a human expert is the way forward in our global fight against TB. In the future, AI-powered tools like GENKI will play a major role in democratizing healthcare and making it accessible to millions of people presently deprived of it.