Research overview

Fluorescence imaging is a powerful tool for recording neural activity in vivo with single-cell resolution. During my Ph.D., I work on computational and optical methods for fast fluorescence imaging recording and analysis. 

Fluorescence imaging analysis

Neuroscientists have been collecting lots of fluorescence imaging data from animal brains. While it is easy to manually segment a few neurons from the fluorescence imaging movie, the work could be laborious when one needs to segment hundreds of neurons. To facilitate this task, I developed the automated pipeline VolPy to analyze voltage imaging data using deep learning tools for neuron segmentation and template matching based algorithm for spike extraction. I also developed a real-time analysis pipeline named FIOLA which utilized GPU hardware to massively speed up the real-time analysis speed, important for closed-loop experiments. Besides, I contributed to a popular calcium imaging analysis software named CaImAn.

Cai C, Friedrich J, Singh A, Eybposh MH, Pnevmatikakis EA, Podgorski K, Giovannucci A. VolPy:automated and scalable analysis pipelines for voltage imaging datasets. PLOS Comp Bio. 2021.

Cai C, Dong CX*, Friedrich J,  Rozsa M,  Pnevmatikakis EA, Giovannucci A. FIOLA: An accelerated pipeline for Fluorescence Imaging OnLine Analysis. (* Equal contributors) Nature Methods. 2023

Compressive microscope

Recently I have been working on optics and microscope design for fast fluorescence imaging recording. In a conventional imaging pipeline, the sensor records an image similar to the targeted scene through a lens. Conventional imaging works well in general when we record a 2D scene. However, much of the information is lost when we want to record a multidimensional scene (let's say a 3D object). The question is that assuming that the scene is compressible, could we use optical coding to manipulate the light field so that the sensor captures more information? The recorded image in the computational imaging pipeline could look very different from the targeted scene but could be used to reconstruct the original 3D object given the optical coding.