Arrhythmia is a medical condition characterized by irregular heartbeats, which, if left untreated, can have life-threatening effects. This project leverages machine learning to detect heart abnormalities using inexpensive electrocardiogram (ECG) data. Our model can classify ECG data as usual or one of four abnormal heartbeats with a false negative rate of only 0.87%. Additionally, our model’s predictions for myocardial activation times are, on average, only 2 ms off. These results suggest that machine learning is a promising, cost-effective solution for arrhythmia detection.
Heart disease is the leading cause of death in the United States. Arrhythmias are a result of underlying heart problems and classifying them can provide life-saving care. A common method of getting heartbeat data is a standard 12 lead ECG, where electric signals from the heart are measured. We aim to use these signals to predict when specific parts of the heart activate.
Here is the task instruction:
- Task 1: Perform binary classification, normal vs abnormal
- Task 2: Further break down to multi-class, identify five categories of heartbeats
- Task 3: Predict the active time of each lead in the heart
- Task 4: Predict whole time sequences curve for heart
For detailed info, refer to the instructions.
- Clone this repo recursively
git clone --recurse-submodules <link for this repo>
-
Set up Dataset
For Task1&2, Dataset can be found on here.
For Task 3&4, Download the dataset by running the following code
source download_intracardiac_dataset.sh -
After set up, run each notebook in order.
-
To run the deep learning model(task 3 and beyond), 32GB of RAM or a GPU is required in order to run.