Automatic assessment of heart sounds applied to telemedicine

Important results from the project

The aim of the project was to improve the technology for remote diagnosis and follow up of patients by using electronic stethoscopes. The goal was to develop and evaluate AI-based methods for classification of heart sound that can measure the signal quality of the recording and finally diagnose the heart as healthy or detect pathological murmurs. This collaboration project has strengthened the AI competence of the project group, but also increased the awareness among our clients and users at Region Västerbotten about the possibilities with this technology.

Expected long term effects

The project result is gained knowledge in AI in general and specifically in signal analysis with machine learning, and a deeper understanding of how to use it in medical applications. We developed a prototype that will be valuable in future discussions and evaluations together with health care personal. This is a summary of the results: - Methods for automatic assessment of signal quality in heart sound recordings. - Methods for classification of heart sound as healthy or pathological. - Prototype showing sound waveform, spectrogram and decision support.

Approach and implementation

The project has progressed according to the original plan with four work packages. First step was literature research and a workshop with AI experts to succeed with algorithm design. In the user study we interviewed medical experts. Most of the work was during development and design of methods for decision support and user interface. The final step focused on evaluating the prototype in a workshop with end users and usability tests. Additionally, we studied a course about deep convolutional networks, participated in the conference Intelligent Health AI and communicated our results.