Date of Award
Master of Science (MS)
Bo D. Liang
The application of machine learning techniques in microscopic image restoration has shown superior performance. However, the development of such techniques has been hindered by the demand for large datasets and the lack of ground truth. To address these challenges, this study introduces a computer simulation model that accurately captures the neural anatomic volume, fluorescence light transportation within the tissue volume, and the photon collection process of microscopic imaging sensors. The primary goal of this simulation is to generate realistic image data for training and validating machine learning models. One notable aspect of this study is the incorporation of a machine learning denoiser into the simulation, which accelerates the computational efficiency of the entire process. By reducing noise levels in the generated images, the denoiser significantly enhances the simulation's performance, allowing for faster and more accurate modeling and analysis of microscopy images. This approach addresses the limitations of data availability and ground truth annotation, offering a practical and efficient solution for microscopic image restoration. The integration of a machine learning denoiser within the simulation significantly accelerates the overall simulation process, while improving the quality of the generated images. This advancement opens new possibilities for training and validating machine learning models in microscopic image restoration, overcoming the challenges of large datasets and the lack of ground truth.
Ling, Cecilia, "Computational Framework For Neuro-Optics Simulation And Deep Learning Denoising" (2023). Theses and Dissertations. 5312.