Projects

Predicting Readmissions from Clinical Data on the Johns Hopkins Prediction Analytics Platform (PMAP)

This project aims to predict readmissions that occur 1-30 days after a Congestive Heart Failure (CHF) - related ICU hospital stay. Currently, I am performing data cleaning, aggregation, and feature engineering processes and anticipate moving on to feature reduction and machine learning at March.

1. Determing the Final Cohort

This notebook first identifies our targeted cohort (CHF-related ICU hospital stays) then generates labels for each hospital stay. The below diagram shows a summary of how the final cohort is extracted out of all of the hospital stays in our dataset.

2. Demographic Features Engineering

This notebook reads in patient demographic information, then generates demographic features including age, gender, and race.

3. Comorbidity Features Engineering

This notebook reads in diagnosis information (ICD-10 codes) from hospital billing and diagnosis data, performs data cleaning procedures, then finally generates the Elixhauser Comorbidity Index.

4. Labs Features Engineering

This notebook reads in each lab entry that occurs within each hospital stay, performs data cleaning procedures, then generates statistical features for numerical labs and binary flags for categorical labs.

5. Meds Features Engineering

This notebook reads in each medication entry that was administered within each hospital stay, performs data cleaning procedures, then generates features including dosages per time per patient weight (if applicable), and binary flags for each medication class.

A Sleep Stage Detection Algorithm based on Recorded Physiological Signals from Smartwatch Sensors for Parkinson᾽s Disease (PD)

This is an algorithm I developed as a Jr. Biomedical Data Scientist at Caduhammer, a startup company in Taiwan that specializes in wearable device algorithms. The algorithm detects Rapid Eye Movement Sleep Behavioral Disorder (RBD) for PD patients and was then contracted by ASUS for their Vivowatch afterwards.

An ECoG-based Brain-Computer Interface (BCI) that uses Deep Learning to Decode Fine Finger Movements

This is my undergraduate research project focusing on developing brain computer interface applications. The project was awarded a $2,000 Undergraduate Research Grant from the National Science and Technology Council.