Stretching Data for MDD Diagnosis

This piece, by Onno Berkan, was published on 02/25/25. The original text, by Wang et al., was published by IEEE Xplore on 01/31/24.

This Zhejiang University study presents DiffMDD, a new artificial intelligence system designed to diagnose Major Depressive Disorder (MDD) using brain wave recordings (EEG). MDD affects approximately 350 million people worldwide, making it one of the most widespread disorders out there. An early diagnosis can lead to treatment and the eventual saving of lives, which then makes the goal of this project all the more important.

The study addresses two major challenges in using EEG for depression diagnosis. First, EEG recordings typically contain a lot of unwanted noise from various sources like environmental interference, eye movements, and heartbeat artifacts. Second, it's difficult to collect large amounts of EEG data due to privacy concerns (healthcare), issues in getting enough people to participate, and the sheer amount of time needed to get sufficient data.

The researchers developed a three-step approach to overcome these challenges. The first step involves deliberately adding controlled noise to the EEG data to help the system learn to identify important patterns even in noisy signals. The second step uses a complicated technique called "reverse diffusion" to generate additional high-quality EEG samples, helping to overcome the limited data problem. The final step involves retraining the system using both the original and newly generated data to make accurate diagnoses.

One interesting finding is that people with MDD show different brain activity patterns compared to healthy individuals. For example, MDD patients typically exhibit lower brain activity levels, with their EEG signals showing less frequent and less dramatic fluctuations. The system learns to recognize these subtle differences to make its diagnoses.

The research team tested their system on two public datasets: Mumtaz2016, which included 34 MDD patients and 30 healthy individuals, and Arizona2020, which contained data from 23 MDD patients and 19 healthy subjects. All participants were in a resting state during EEG recording, with measurements taken from 19 specific points on the scalp.

The results showed that DiffMDD achieved state-of-the-art performance in diagnosing MDD. The system demonstrated particular strength in achieving balanced accuracy, meaning it was equally good at identifying both MDD patients and healthy individuals. This is especially important in medical diagnostics, where both false positives and false negatives can have serious consequences.

What makes this system particularly innovative is its ability to generate new, high-quality EEG data while maintaining the distinctive characteristics of both healthy and MDD patients. For instance, in both the original and generated data, healthy subjects' brain activity was more evenly distributed across different frequency bands, while MDD patients' brain activity was concentrated in lower frequency bands, indicating less active brain states.

The researchers conclude that their approach effectively addresses the challenges of noisy data and limited sample sizes in EEG-based MDD diagnosis, making it a promising tool for clinical applications. However, they acknowledge that future work should include testing on larger datasets and further evaluation in practical hospital settings.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!