Inspiration
After working as an intern at a local Family Medicine clinic for two years, I often ran EKG scans on elderly patients to assess overall heart health. I realized that this process was inefficient, as patients had to be connected to the machine and maintain an uncomfortable position until the doctor returned to review the scans after visiting other patients. I wanted to create a tool that could identify high-risk patients with reduced reliance on the physician in order to streamline this process, reducing physical strain on elderly patients and improving the accuracy of physician diagnoses.
Function
This tool analyzes individual patient EKG scans and highlights any regions/heartbeats that resemble common heart irregularities at a high confidence level. This model utilizes a 1D convolutional neural network, trained on hundreds of thousands of heartbeat samples from the MIT-BIH Arrhythmia Database.
Relevant Skills
PyTorch, NeuroKit2, WFDB, Streamlit, Plotly, OpenCV, scikit-learn, imbalanced-learn