Examples of code from my prior projects
May 2020
Used TensorFlow to create a deep learning model for electrocardiogram (ECG) analysis based on WaveNet, a deep learning model for audio signals. Diagram from creators of WaveNet:
I created one version to take in one-dimensional input - the ECG amplitudes for a single lead. Also created another version to take in two-dimensional input, with each lead of a 12-lead ECG along the second dimension.
The model was used for various predictions, including estimation of age, which was also performed by a group at the Mayo Clinic using a convolutional neural network. Their CNN was trained on ~500,000 ECGs. Their results:
I created a deep learning model that emulated the model architecture described in the Mayo Clinic paper. Below is the comparison of performance after using the same set of ~13,000 hospital ECGs to train my WaveNet-based model (35,500 parameters trained) and the emulated Mayo Clinic model (197,000 parameters trained).
July 2021
Used PyTorch to create a UNet model to perform segmentation of the mitral valve on echocardiogram (cardiac ultrasound) videos. Included options for data augmentation in the scripts. Resulting model segmented mitral valve on each frame of test videos.
May 2020
Published peer-reviewed paper on coronary artery disease (CAD) as a risk factor for heart failure with preserved ejection fraction (HFpEF). Used Stata to analyze ARIC cohort data, using epidemiology and biostatistics skills. Project involved extensive data preparation steps to create the appropriate variables.
Miscellaneous functions used across Python projects
January 2021
Used ImageJ's Java-like programming language to set up a video annotation workflow. This enabled a more robust team collaboration and feedback system than was possible with the many annotation software vendors I met with.
December 2019
Used regex within Python to extract infromation from cardiac catheterization procedure reports.
December 2023
Performed web scraping of FDA 510k submissions to identify those pertaining to cardiology AI.
June 2005
Joined a neuroscience lab in college that analyzed and visualized EEG data in Excel. Learned Excel VBA from a book and scripted a macro to automate these steps, saving several hours of work per week.
October 2023
Created slide deck for a nontechnical audience assessing AI-based healthcare tools. I plan to convert this to a series of Medium posts to help improve healthcare AI literacy.