B.S.  Electrical Engineering	Tsinghua Univ.	1999
M.S.  Electrical Engineering	IACAS (China)	2002

The human brain is the most complicated system and only a little is known about its organization and way of working, although we are using it every time on everything we do. Invasive intracranial study provides some information about the brain function, but it is limited to a small group of patients. To know how a normal brain processes information and makes response, we have to use non-invasive modalities. fMRI is widely used for its good spatial resolution when we look at the hemodynamic brain activities. Unfortunately, its temporal resolution is poor. EEG, on the other side, shows very good temporal resolution whereas its spatial resolution is not so satisfying. My research now focuses on EEG/fMRI study of brain function, especially about attention and learning.

Research Projects:

1. Change-blindness

Reference

Figure 1

Figure 2

Take a look at these figures. Can you see any difference between them? Maybe can not at the first glance. In Figure 1, one of the submarines changed its color and in Figure 2, another submarine changed its direction compared with the reference figure. Study shows that people are quite poor at noticing changes to objects in photographs or movies from one instant to the next. This is so-called "change-blindness". Attention is needed for the change to be detected.

The above figures are coming from a game we designed to see how a primary task influence the subject's attention and ability to detect the changes. In this game, the subjects need to control the boat to avoid those torpedoes shooting from the underwater submarines (primary task) and at the same time detect the color/direction changes of the submarines (secondary task). We also want to see the response difference between change-blindness and change-detection in EEG.

We see that when the primary task is very difficult (more torpedoes within a unit time), the probability of change-detection is reduced, whereas the secondary task doesn't modulate the performance of the primary task. For the response after a change is detected, a positive deflection around 300ms after the change onset is observed (Figure 3). For single trial analysis, we got a ROC Az peak value about 0.8 at 300ms after the event onset using logistic regression method (Figure 4).

Figure 3

Figure 4

2. P3 vs. work load

P3 is a prominent positive event-related potential (ERP) component happening around 300ms after a target. It is related to target detection and categorization. It is also an indication of the brain's work load. Study shows that when workload is heavier, P3 amplitude will decrease, as follows.

In this project, we used a two scenarios for task: (1) Hard: avoid torpedoes (high rate) and detect visual change. (2) Easy: avoid torpedoes (low rate, 1/3 of high rate). Tones with 1300Hz were used to elicit the P3. The timeline are shown in Figure 5. From Figure 6 we see that logistic regression is able to classify hard conditions from easy ones.

Figure 5

Figure 6