MediCats

Problem: Healthcare systems in Europe are overstrained; there are not enough doctors to treat all the patients in time, while on the other hand, about 70% of medical visits that happen, are unnecessary. Doctors’ and nurses’ time is not optimally distributed towards patients that would need treatment. Nowadays, there are huge amount of data available, such as past medical records, data about activity as well as other data from wearables (such as ECGs). The potential of this data is currently unused.

Solution: We are introducing MediBee - a health monitoring assistant for patients and doctors. It is a mobile app, that uses artificial intelligence to support patient's health journey. We have trained a ML model that uses convolutional NNs to detect irregularities in ECGs and trigger alert in the app. After the user provides additional information (if this is not yet available via wearables), the data is sent to a doctor, who has a final decision whether the person should come in for a checkup.

Impact: 70% of doctors’ visits are unnecessary, which causes huge costs for the healthcare system as well as lack of resources for people who really need medical assistance. By reducing the number of redundant visits by monitoring patients' data with help of AI, waiting queues and doctors’ overtime would decrease. Additionally, we could diagnose people earlier, which would dramatically increase their quality of life. MediBee aims to provide doctors’ assistance to every patient at the right time point.

Feasibility & financials: CNNs have shown good results in identifying irregular heart rhythms, whereas other research shows that aortic stenosis can be identified from ECG data from wearables. Results that we have obtained are also promising (0.98 precision, 0.99 recall on train data). We would propose to implement such a project in an iterative way - to get to the first MVP, investment of 48k eur would be needed. The continuous improvement and collaboration with doctors and domain experts is essential for success.

What is new about your solution?: MediBee combines multiple data sources to provide reliable and accurate detection of anomalies. Since some data is always missing, in case of detected irregularities, patient is asked via chatbot to provide additional info, that would improve the prediction. Our app is not aiming to replace doctors, but to empower them and take off some of their workload, when it comes to routine checkups. On the other hand we put a lot of emphasis on clean UX- we want users to love the app and use it regularly.

What you have built at the hackathon - text explanation + code (e.g. GitHub link): We have done exploratory data analysis and data cleaning of all 3 datasets. Main efforts went into analysis of 6 lead ECG data originating from wearables. We preprocessed the data, split it to train and validation set and built a ML model with convolutional neural networks that detects whether ECG is “bad” or “healthy” one. We have also prepared a clickable prototype of how would the mobile app for patients look like, as well as design and prototype of a doctors dashboard. (see project link)

What you had before the hackathon, please mention open source as well: Nothing, except for motivation and enthusiasm :) We came to the event open minded and ready for solving the challenges that we were not even aware of. When implementing technical solutions, we have used some existing standard libraries that are used for data handling (pandas, scikit-learn, numpy) and framework keras (based on tensorflow), which can be used to build a neural network model.

What comes next and what you wish to achieve: MediBee for ECG monitoring is just the beginning-our vision is to become a versatile medical assistant. There is a lot of patients’ data available, the potential of combining various datasources is huge and our concept can be applied to many medical domains. We think collaboration is of essence: technology working together with and for people, patients and doctors collaborating, and lastly, our team building this product together with YOU, who can directly feel the pain points of current system!

Video: https://www.youtube.com/embed/w37jSBhoN8g