03/19/2026
By Danielle Fretwell

The Francis College of Engineering, Department of Electrical and Computer Engineering, invites you to attend a Doctoral Dissertation defense by Shuo Han on: "Generative Model-based Cardiac Computed Tomography Reconstruction from Incomplete Projection Data."

Candidate Name: Shuo Han
Degree: Doctoral
Defense Date: Wednesday, April 1, 2026
Time: 2 - 3:30 p.m.
Location: Ball Hall 302

Committee:

  • Advisor: Hengyong Yu, Professor, Electrical and Computer Engineering, UML
  • Yan Luo, Professor, Electrical and Computer Engineering, UML
  • Seung Woo Son, Associate Professor, Electrical and Computer Engineering, UML

Abstract:
Computed tomography (CT) has become an essential imaging modality in medical diagnosis, offering detailed anatomical insights for various clinical applications. Specifically, cardiac CT has emerged as particularly valuable for diagnosing and managing cardiovascular diseases due to its ability to visualize the heart's anatomy and function in detail. However, cardiac CT requires high temporal resolution to effectively capture the rapid motion of the beating heart. Limited-angle data acquisitions help achieve this by shortening scan times, but this technique inherently results in incomplete projection data, leading to substantial image degradation.

Leveraging generative diffusion models, this dissertation investigates novel reconstruction methods to overcome challenges associated with incomplete projection data in cardiac CT. First, a physics-informed score-based diffusion model (PSDM) is proposed, integrating data-driven priors from diffusion models with model-driven priors derived from the primal-dual hybrid gradient (PDHG) algorithm and Fourier fusion techniques. Second, an optimized 3D diffusion based reconstruction method is developed, combining regularized derivative least squares (RDLS) algorithms and denoising diffusion implicit models, significantly reducing computational time while maintaining high-quality image reconstruction. Third, we propose a rectified-flow framework for single-step limited-angle cardiac CT reconstruction. The model learns a deterministic mapping from artifact-corrupted FBP reconstructions to fully sampled ground truth using a 2.5D slice-context strategy to leverage neighboring slices. A reconstruction-aware spectral loss further enforces frequency-domain fidelity, improving missing-wedge recovery and structural consistency while substantially reducing inference time.

Comprehensive numerical simulations and real-data experiments confirm the effectiveness of the proposed generative diffusion approaches in reconstructing high-fidelity cardiac CT images, substantially mitigating reconstruction artifacts and enhancing temporal resolution. These methodologies represent significant improvements in cardiac CT imaging, ultimately contributing to improved diagnostic accuracy and clinical outcomes.