Electrocardiograms (ECGs) are an essential first line tool in the diagnosis of many different types of heart conditions. However, a complete set of ECG contains 12 leads of signals, making the ECG collection process time-consuming. Moreover, one study found that only 10% of participants (doctors, nurses, and cardiac technicians) correctly applied all the measuring electrodes. Misplaced chest electrodes could significantly affect the quality of signals and increase the chance of missing diagnoses. This motivates us to seek a technical solution that can reduce the time and complexity of ECG collection and the risk of misdiagnosis.
In this project, we investigate various machine learning strategies proposed by groups of biomedical researchers which are used to reconstruct the 12-lead ECG signals given only 2 or 3 leads of signals as input. Developers will go through a complete research process, including comprehending academic papers, developing machine learning models, preprocessing and training data, comparing with other models and making improvements on the current machine learning frameworks, etc. We will first reproduce the existing models, and then try to develop architectures that could achieve better performances. Experience in using tools such as PyTorch/TensorFlow, pandas, NumPy is an asset, but all the machine learning lovers are welcome!
We are looking for developers with experience in research and/or computer vision to join us. Email firstname.lastname@example.org or DM “Yan Zhu” in the UTMIST Discord if you’re interested!