Research

Our group is interested in developing next-generation optical imaging and computational methods that will enable noninvasive (label-free), deeper, faster, and richer (more contrast) visualization of biosystems.

Current research:

Advanced microscopy
Microscopy and AI
Light source
Wavefront shaping

Advanced microscopy

Understanding of complex diseases, therapeutic effects, and biological functions require new imaging tools with minimal biological disruption, multiplexed molecular contrast in high volumes and high resolution. We are developing next-generation microscopy methods that will enable noninvasive (label-free), deeper, faster, and richer (more contrast) visualization of biosystems.

Microscopy and AI

Besides optical methods, algorithms can play an important role in this pursuit as well. Previous work by others and us have shown AI can efficiently retrieve biologically relevant information from multi-dimensional dataset. How to reduce the demand of training data to make this easily accessible for more biology labs? How to use the real-time nature of AI algorithms to strategies image acquisition in real time?  We are developing AI-assisted microscopy to retrieve more information with less cost in time, instrument, and phototoxicity.

Light source

Advances in nonlinear optical imaging have been highly correlated with advances in ultrafast light sources. With the increasing demand of multiplexing, imaging depth, and speed, there is an urgent need for more versatile and flexible light sources. By leveraging nonlinear fiber optics and machine learning, our lab is exploring new ways of generating tunable sources for advanced bioimaging.

Wavefront shaping

The capability to engineer the wavefront enables us to shape the light in 3D volume, to generate holography, and to overcome scattering effects. However, the speed and accuracy of wavefront shaping is a longstanding problem due to hardware constraints. We are developing instrumentations and algorithms to model and solve this ill-posed inverse problem, which will advance our capability to shape the light for applications such as computer-generated holography, deep tissue imaging, and endoscopy.

 

Previous research:

Development of Label-free multiphoton microscopy
Label-free in vivo microscopy
AI-driven translation into clinic
Raman spectroscopy and imaging

Label-free multiphoton microscopy

Using fiber supercontinuum and pulse shaping, we developed a multiphoton imaging platform that achieves real-time label-free single-shot functional and structural optical imaging for the first time, and enables non-perturbative, faster, and richer visualization of living systems.

Lable-free in vivo microscopy

Because of the nonlinear dependence on incident power and the longer excitation wavelength, multiphoton imaging achieves deeper imaging with less photobleaching artifacts compared to one-photon imaging. This opens windows to in vivo imaging of the dynamic living biosystems. Label-free multiphoton microscopy makes the technique even more attractive by the efficient excitation and detection of the intrinsic molecular contrasts.

By utilizing the intrinsic metabolic signatures of biological molecules, we were able to directly visualize and characterize extracellular vesicles in living animals and fresh human tissues with breast cancer.

AI-driven translation into clinic

The high (~30%) re-operation rate for breast conserving surgeries arise from the failure of detecting positive margins at the time of surgery. This risk can be significantly reduced by a combined framework of real-time virtual histopathology and a deep-learning-based classification algorithm.

Recent advances in microscopy, along with advances in artificial intelligence (AI), present an exciting opportunity to innovate microscopy analysis and instrumentation. Here is an example of how we use deep learning to “pseudo label” the label-free images.

Raman spectroscopy and coherent Raman imaging

Raman spectroscopy provides noninvasive fingerprints of chemical compositions at molecular level. Coherent Raman scattering makes it possible to extend this chemical profiling capacity to broader imaging applications that require high spatiotemporal resolution on top of the chemical resolution, e.g. drug tracking in living cells as shown in the figure below. Signal throughput has been a longstanding limit to further expand this technique to real-life biomedical applications. We aim to develop methods that can address this issue by efficiently acquiring and reconstructing the hyperspectral dataset.