ADHD Research

Identifying ADHD Biotypes using Machine Learning

Overview

This is a project backed by a Stanford HAI seed research grant, aiming to aid ADHD diagnosis and biotype identification, a vastly unexplored territory usually clinically determined instead of data-driven. Our insight relies on applying machine learning on electrodermal activity (EDA) signals along with other bio-signals to find patterns, and if successful, this would make diagnosis accessible through simple wrist-worn sensors. I am a research assistant focusing on engineering a data pipeline for feature extractions and brainstorming unsupervised learning methods. 
  • Timeline
  • Sep 2022 - Present
  • Mentor
  • James Landay
    Nava Haghighi
  • Skills
  • Research
    Data Engineering
    Data Analysis
    Signal Processing
  • Tools
  • Pandas
    Scikit-learn
    R/RStudio
    Matplotlib

Given a small amount of incredibly rich, dense, and long time series data, how can we best extract the most important features and factors?

How can we automate the analysis pipeline and scale for the next round of the study? How can we organize and apply quality control to the data we have now?

For me, I wanted to get a taste of formal long-term, interdisciplinary research, while learning more data engineering and analysis tools and exploring bio-data and signal processing for the first time.

Action

Data Pipeline

I’m engineering a data pipeline that cleans bio-signals (motion, EDA, temperature) from ~10 billion time series data among 40 participants, extracting and visualizing 120+ features from biosignals using Pandas, Scipy, and Matplotlib

Motion Artifacts

I’m testing a novel method of motion feature extraction for tri-axis accelerometer data using the R package GGIR, comparing it against other previous methods for E4 sensors

Feature Extraction

I am combining all of sensor data to extract features from EDA signals. This includes calculating (using different techniques in Neural-kit2, Biosppy, and cvxEDA) SCR, SCL, mean amplitude, motion vs motionless, sleep, and more.

Outcome

Since this is a long-term research project involving multiple rounds of studies and analysis, my work is only the beginning to set up some ground work. Nonetheless, this was the most progress made in a while as commented by the lab PI.  

Please reach out if you're interested in the details!

Reflections & Takeaways

Don’t get afraid to get your hands dirty! Having a consistent ritual and system is very important to get through some tedious work. Communication is key. Not all AI research is done in half a year. It's also important to have self compassion when stuck.

Thanks for stopping by!

Ryan Lian ©2023

OverviewChallengeActionoutcomeReflection