My research

Analyze the patterns of whole brain activities

Recent techniques in microscopy and image processing have allowed scientists to monitor and extract the whole brain activities (WBA) at the single neuron level, while the underlying patterns in the whole brain activities have not been well understood. To solve this problem, I am studying  WBA in C. elegans, which has a simple brain with only 302 neurons and all of the physical connections between neurons have been known. A summary of the whole brain activity studies can be found in our previous review (Wen and Kimura, JJAP, 2020).

Efficient manual correction software for segmenting 3D cells in EM images

While deep learning techniques have substantially improved the accuracy of autamated cell segmentation. It still generates lots of mistakes. Correcting these mistakes can be quite time-consuming, especially for the EM images with large sizes and huge number of cells. We developed Seg2Link, which is a software to help users easily correct the mistakes in the automated segmentation results and thus quickly obtain the accurate 3D structure of cell populations.

Source code and instructions for use Seg2Link: https://github.com/WenChentao/Seg2Link

Track cells in 3D time lapse images in deforming/moving organs

To extract cellular activities from various organs, an algorithm that can flexibly segment and track the moving cells in different conditions is required. I have developed a deep-learning-based algorithm, 3DeeCellTracker, which can accurately track cells in 3D time-lapse images, on a desktop PC with a GPU. 3DeeCellTracker successfully tracked cells in various conditions/organs such as the cells in the tumor spheroid, ("straightened") freely moving worm, and the beating fish heart. (Wen et al. eLife, 2021; Voleti, et al. Nat. Methods, 2019) 

Python package and Jupyter notebooks: https://github.com/WenChentao/3DeeCellTracker 

The neural mechanism for updating reward prediction

To survive in an ever-changing environment, an animal's brain needs to predict the reward (such as food) received in the future, and update the prediction when the reward changed. Previous studies have strongly implied the dopaminergic neurons in the midbrain code the error signal which plays a key role in updating the predictions. However, how this error signal is computed in the brain has been unknown. By comparing the recorded neuronal activities and the learning theory, I found that two groups of neurons in the striatum may code the reward signal and the prediction signal, respectively, whose difference is the error signal. These results together with our findings of striatum-midbrain connections suggested a potential neural mechanism for computing the error signals important for learning. (Wen et al. Front. Neurosci. 2016)

My other interests 

Except for the above studies, I'm also interested in the neural substrates of consciousness.