Find the Fib

Inspiration

Atrial fibrillation is the commonest cardiac arrhythmia, and one of the comments cardiac problems. It affects around 4% of the elderly population above sixty, and goes up to 8% from 80 years and older. It leads to a 5-fold increased risk of strokes. Approximately 70% of AF related strokes can be avoided with adequate diagnosis and treatment. The difficulty with AF is making the diagnosis, as up to 40% of patients are asymptomatic, and even in those with symptoms, there is a low detection rate with single ecg readings. Holter ECGs are expensive and often our older population do not have access to this technology. We were inspired by the idea of creating a product which could increase our ability to detect diseases which has such a big impact on the lives of our elderly population.

What it does

We designed a device and a service that work together to facilitate continuous monitoring of the Electrocardiogram of patients with similar accuracy and lower costs than the current technology. Moreover, your software is scalable to devices with small form factors, so there is potential for a wearable ECG classifier product to be developed.

How we built it

We started by designing our hardware device on paper, and migrated it to Autodesk. Due to time and financial contraints, we were unable to physically build our device. Our software is inspired by the DeepECG model (https://github.com/ismorphism/DeepECG) and was trained on the Physionet Challenge 2017 dataset (https://archive.physionet.org/pn3/challenge/2017/). We used an AWS instance to train the model and then hosted a web UI on Google Cloud to facilitate easy user interaction.

Challenges we ran into

Our main challenges were related to the time constraints of prototyping within a 36 hour window. We also imposed some financial constraints and form factor constraints on our custom device. As with a lot of machine learning projects, we faced memory and compute power shortages during the training phase, and had to improvise when we ran out of our Google Cloud CPU quota. Our backend and frontend were also notoriously hard to connect to each other due to a few network issues and related problems.

Accomplishments that we're proud of

Firstly, we overcame every challenge, bug and failed build thrown at us over this 36 hour period to create something entirely new. Second, each of us was not only able to use our expertise in to benefit the team, but we were able to learn from each other and expand our interests and knowledge. Finally, we all demonstrated our dedication and commitment to the project by pulling an all-nighter to work!

What we learned

The most important thing we learned was how to work together, ideate, problem solve and effectively communicate with each other. Each of us also expanded our own interests and knowledge through the workshops and each other's experience.

What's next for Find the Fib

We want to expand on the diagnostic capability from exclusively Atrial Fibrillation to other cardiac anomalies such as other arrhythmias and heart lesions and even myocardial infarction. We also plan to scale our device down to a wearable form factor and incorporate wireless and bluetooth capabilities. Finally, we recognize that this technology can be applied to various populations other than elderly patients, and we want to adapt our devices and servcies in any way necessary to facilitate easier diagnoses and treatment for everyone.