The low cost and high availability of lung sounds makes them important in medical diagnosis. However, the subjective interpretation, and hence detection of abnormal sounds has worryingly high variance. Currently there is also an increasing interest to use lung sounds in smart medical and home monitoring devices. However, lung sound analysis methods have mostly been developed and tested using small, and unrealistic, datasets and have therefore been of little practical use. There is therefore a need for better tools to train future doctors and nurses. These tools can also be used by non-professionals. We are seeing more and more smart devices, wearables, and apps targeted for self-monitoring of various health metrics, including smart stethoscopes. These are also integral parts of the mobile platforms provided by Apple and Google. In Tromsøundersøkelen we have collected the biggest epidemiological lung sound dataset in the world. We have had leading lung sound experts in the world annotate the recorded sounds. We have therefore been the first research group in the world that could develop deep learning methods for lung sound analysis using a big realistic dataset. We have already identified training and monitoring tools for our methods. We have already implemented prototype tools utilizing the analysis results, and our lung sounds experts are already evaluating these. Our solution gives our users the needed skills, without the need for human trainers. We know that there is interest to use such tools at the University of Tromsø (UiT). We want to establish a startup. We will build and deliver a cloud-based platform for lung sound analysis, and implement a training tool based on our prototype. We plan to test the tool in a pilot study at the nurse education at UiT, and then sell subscriptions to other universities at a global scale. We will port the tool to mobile platforms for use with Bluetooth stethoscope, and sell it through mobile app stores.
Project leader: Johan Fredrik Eggen Ravn
Institution: MEDSENSIO AS