AI based model for prediction of undiagnosed Atrial Fibrillation

Important results from the project

The main objective of the project was to investigate the feasibility, and performance of an AI-based algorithm to distinguish high-risk patients for having an undiagnosed paroxysmal atrial fibrillation from low-risk patients. The goal was to exceed the performance of the corresponding NT-proBNP blood test. The project has proven that AI-trained models can be used for the purpose. In the project, Zenicor has also acquired essential competence and experience in AI through the collaboration with the specialists at Modulai AB.

Expected long term effects

The project has shown that an AI-powered algorithm can be successfully used to identify high-risk patients for having a paroxysmal atrial fibrillation with a performance that exceeds previously used blood samples. In a future clinical scenario, the model could be used to easily and cost-effectively distinguish the patients who have the greatest need for an in-depth screening for atrial fibrillation through a rapid ECG test at a health center or pharmacy. The project has proven the benefit of using AI for the purpose and give incentives for further development.

Approach and implementation

The project was carried out in collaboration with academia and a consulting company focusing on Artificial Intelligence. Through the collaboration with researchers at Karolinska Institutet and Cambridge University, the project was able to gain access to annotated raw ECG data and ensure clinical benefit and relevance of the model. By hiring experienced specialists in AI, the project was also able to effectively identify, compare and validate different methods in the art of AI.