Arya Chowdhury Mugdha
Ph.D. PreliminaryNov 14, 2023, 1:00 pm - 3:00 pm
Phoenix Design Studio, Scott Building
Study of Mitochondrial Heterogeneity in live fibroblasts using label-free optical microscopy and unsupervised neural networks
Abstract: The focus of this thesis is to study mitochondrial heterogeneity in cultured human fibroblasts utilizing a transient absorption laser scanning system developed in our lab. Mitochondrial heterogeneity is a condition that occurs when a cell or tissue contains both healthy and diseased mitochondrial DNA. The quantification of healthy and diseased mitochondria is a crucial factor in identifying mitochondrial disease. We hypothesize that transient absorption microscopy, which utilizes femtosecond laser pulses to measure sub pico-second excited state relaxation mechanisms of electron transport chain (ETC) hemeproteins, can quantify intracellular mitochondrial heterogeneity. As proof-of-concept, we demonstrate that excited state absorption processes in live cultured human fibroblasts can be investigated using transient absorption technique.
Another aspect of this thesis is to develop unsupervised machine learning based algorithms to deal with the interpretability of transient absorption data. Time resolved microscopy techniques like fluorescence lifetime or transient absorption can have low signal-to-noise ratio or non-orthogonal exponentially decaying spectral signatures. These characteristics are important to describing the physics behind image contrast. We demonstrate that an autoencoder based on convolutional neural networks can unmix the spectral signatures present in transient absorption images collected from muscle fibers. We also investigate other state-of-the-art deep learning algorithms such as Noise2noise for signal-to-noise ratio improvements in some of our fibroblast images collected using transient absorption technique.
Another aspect of this thesis is to develop unsupervised machine learning based algorithms to deal with the interpretability of transient absorption data. Time resolved microscopy techniques like fluorescence lifetime or transient absorption can have low signal-to-noise ratio or non-orthogonal exponentially decaying spectral signatures. These characteristics are important to describing the physics behind image contrast. We demonstrate that an autoencoder based on convolutional neural networks can unmix the spectral signatures present in transient absorption images collected from muscle fibers. We also investigate other state-of-the-art deep learning algorithms such as Noise2noise for signal-to-noise ratio improvements in some of our fibroblast images collected using transient absorption technique.
Adviser: Jesse Wilson
Co-Adviser: N/A
Non-ECE Member: Stuart Tobet, Biomedical Engineering
Member 3: Randy Bartels, Electrical and Computer Engineering
Addional Members: Kevin Lear, Electrical and Computer Engineering
Co-Adviser: N/A
Non-ECE Member: Stuart Tobet, Biomedical Engineering
Member 3: Randy Bartels, Electrical and Computer Engineering
Addional Members: Kevin Lear, Electrical and Computer Engineering
Publications:
A.C. Mugdha and J.W. Wilson, Blind hyperspectral unmixing of pump-probe images with maximum likelihood selection of a convolutional network from an ensemble of models, SPIE Photonics West Conference, San Francisco, CA, 2023,
J.W. Wilson, E. Wang, A. Mugdha, L. Whitcomb, A. Chicco, Label-free visible wavelength transient absorption imaging: mitochondrial redox contrast and unsupervised spectral learning, Ultrafast Nonlinear Imaging and Spectroscopy X, PC122280P, 2022
E.E. Flater, A.C. Mugdha, S. Gupta, W. A. Hudson, A. A. Fahrenkamp, Jason P Killgore, J. W. Wilson, Error estimation and enhanced stiffness sensitivity in contact resonance force microscopy with a multiple arbitrary frequency lock-in amplifier (MAFLIA), Measurement Science and Technology, 2020
A.C. Mugdha and J.W. Wilson, Machine learning approach to synthesizing multiphoton microscopic images from reflectance confocal, SPIE Photonics West Conference, San Francisco, CA, 2019
A.C. Mugdha and J.W. Wilson, Blind hyperspectral unmixing of pump-probe images with maximum likelihood selection of a convolutional network from an ensemble of models, SPIE Photonics West Conference, San Francisco, CA, 2023,
J.W. Wilson, E. Wang, A. Mugdha, L. Whitcomb, A. Chicco, Label-free visible wavelength transient absorption imaging: mitochondrial redox contrast and unsupervised spectral learning, Ultrafast Nonlinear Imaging and Spectroscopy X, PC122280P, 2022
E.E. Flater, A.C. Mugdha, S. Gupta, W. A. Hudson, A. A. Fahrenkamp, Jason P Killgore, J. W. Wilson, Error estimation and enhanced stiffness sensitivity in contact resonance force microscopy with a multiple arbitrary frequency lock-in amplifier (MAFLIA), Measurement Science and Technology, 2020
A.C. Mugdha and J.W. Wilson, Machine learning approach to synthesizing multiphoton microscopic images from reflectance confocal, SPIE Photonics West Conference, San Francisco, CA, 2019
Program of Study:
ECE-604
ECE-503
ECE-504
ECE-512
ECE-457
ECE-571
CS-545
CS-510
ECE-604
ECE-503
ECE-504
ECE-512
ECE-457
ECE-571
CS-545
CS-510