Research Overview

The main focus of the Computational Biomedical Imaging Group (CBIG) is on the development of theoretical and algorithmic tools for machine learning in the presence of missing data. CBIG's research applies to biomedical imaging, particularly magnetic resonance imaging (MRI).

Our lab is at the forefront of developing cutting-edge methodologies that bridge advanced machine learning techniques with innovative imaging solutions. Our research encompasses a range of interdisciplinary projects aimed at enhancing data analysis and imaging accuracy across various domains.

Model-Based Supervised and Unsupervised Deep Learning: We are pioneering efforts in integrating model-based approaches with deep learning to tackle both supervised and unsupervised learning challenges. By incorporating structured models into deep learning frameworks, we aim to improve performance, interpretability, and generalization in diverse applications.

Smoothness Regularization on Manifolds (STORM): Our STORM project focuses on incorporating smoothness constraints into learning algorithms for data that lie on complex manifolds. This work is dedicated to enhancing model stability and generalization by ensuring solutions are smooth and adhere to the underlying manifold structure.

Deep Structured Low-rank Algorithms for Uncalibrated MRI: We are developing advanced deep learning algorithms that exploit low-rank structures to improve the quality of uncalibrated MRI data. This research aims to reduce imaging artifacts and enhance the clarity of MRI scans by integrating structured priors into the imaging process.

Continuous Domain Compressed Sensing: Our lab is advancing compressed sensing techniques for continuous domains, aiming to refine signal recovery and analysis from incomplete data. This involves creating algorithms that handle continuous signal representations more effectively, thereby improving data acquisition efficiency.

Union of Surfaces Model for Data Living on Manifolds: We propose a novel model that represents complex data on manifolds as a union of surfaces. This approach enhances the modeling and interpretation of high-dimensional data by leveraging its intrinsic geometric structure.

Learned Image Representations for Multidimensional Imaging: Our research focuses on developing learned representations to enhance multidimensional imaging techniques. By automating the learning and representation of complex imaging data, we aim to improve the analysis and interpretation of multidimensional images.

High Resolution Metabolic MRI Using Learned Models: We are pushing the boundaries of metabolic MRI by applying learned models to achieve high-resolution imaging. This project seeks to enhance the detail and diagnostic capability of metabolic imaging, crucial for studying and diagnosing metabolic disorders.

High-Resolution Brain Connectivity Mapping: Our lab is dedicated to developing methods for creating high-resolution brain connectivity maps. This research aims to provide a detailed understanding of brain network interactions and improve the accuracy of connectivity mapping.

Fast Graph Search Algorithms for Fat-Water MRI: We are innovating in the development of fast graph search algorithms to accelerate the processing and analysis of fat-water MRI images. These algorithms are designed to efficiently differentiate between fat and water components, improving the diagnostic and analytical capabilities of MRI.

Through these diverse and interrelated projects, our lab is committed to advancing the state-of-the-art in machine learning and imaging technologies, addressing complex challenges, and driving forward innovations that have a significant impact on healthcare and beyond.