Abstract: Bridging Statistical Decision Theory and Machine Learning for Imaging
Recent advances in computational methods and artificial intelligence (AI) are creating new opportunities for analyzing and optimizing imaging systems. In this talk, I will present recent methodological developments at the intersection of computational image science and machine learning. I will discuss how deep generative models can be used to learn stochastic object models from imaging data and facilitate Bayesian ideal observer computation. I will then introduce a novel framework for extracting task-relevant features from high-dimensional imaging data, enabling efficient observer analysis in a low-dimensional feature space. Finally, I will demonstrate a task-based approach for image-specific fine-tuning of neural network-based denoisers.