ECE PhD Prospectus Defense: Guangrui Ding

ECE PhD Prospectus Defense: Guangrui Ding

Title: Pushing the Imaging Depth of Chemical Imaging by Advanced Instrument and Artificial intelligence (AI^2)

Presenter: Guangrui Ding

Advisor: Professor Ji-Xin Cheng

Chair: Professor Lei Tian

Committee: Professor Ji-Xin Cheng, Professor Darren Roblyer, Professor Lei Tian, Professor Vivek Goyal, Professor Haonan Lin

Google Scholar Link: https://scholar.google.com/citations?hl=en&user=iHRPqbMAAAAJ&view_op=list_works&sortby=pubdate

Abstract: Coherent Raman scattering (CRS) offers powerful, label-free visualization of biomolecule distributions in living systems. However, its imaging depth is constrained by low Raman cross section and multiple scattering. By co-optimizing Advanced Instrumentation and Artificial Intelligence (AI^2), this proposal overcomes these limitations via the AI2 strategy, including advanced shortwave infrared (SWIR) coherent Raman imaging and deep learning-based contrast enhancement.

We will develop a high-power CRS imaging system based on 1 MHz femtosecond non-collinear optical parametric amplifier (NOPA) laser.

Compared with conventional CRS driven by 80-MHz optical parametric oscillators (OPO), NOPA CRS enables two orders of magnitude signal enhancement. To mitigate photodamage and suppress background noise, we introduce extensive pulse chirping, allowing high-peak-power excitation under biosafe conditions. To resolve multiple scattering, we will switch the excitation wavelength from near infrared (NIR) to SWIR region. The stronger Raman signal and weaker scattering together improve the image depth to millimeter scale. On the computational front, we propose a model-based self-supervised deep learning method to reduce noise and enhance contrast. This includes Fourier Shell Correlation (FSC)-assisted Noise2Noise denoising to remove 3D non-identical noise, followed by data-informed point spread function calibration and deconvolution. The method is expected to improve the learning stability and enhance contrast in heterogeneous tissue environments. We will demonstrate the performance through label-free visualization of lipid and protein distribution in 3D tumor spheroids.

In summary, the proposal aims to achieve highly sensitive deep tissue chemical imaging for label-free molecular fingerprinting in complex biological systems.