11/21/2025
By Danielle Fretwell

The Francis College of Engineering, Department of Electrical and Computer Engineering, invites you to attend a Doctoral Dissertation defense by Bahareh Morovati on: "Photon-Counting Computed Tomography Image Reconstruction and Data Correction through Deep Learning Technique."

Candidate Name: Bahareh Morovati
Degree: Doctoral
Defense Date: Monday, December 1, 2025
Time: 2:30-4 p.m.
Location: Ball 302

Committee:

  • Advisor: Hengyong Yu, Professor, Electrical and Computer Engineering, UMass Lowell
  • Yan Luo, Professor, Electrical and Computer Engineering, UMass Lowell
  • Seungwook Son, Associate Professor, Electrical and Computer Engineering, UMass Lowell

Abstract:
Medical imaging plays a vital role in modern healthcare by enabling noninvasive visualization of internal anatomy and physiology for disease diagnosis, treatment planning, and monitoring. Among various modalities such as Magnetic Resonance Imaging (MRI), Ultrasound, and X-ray, Computed Tomography (CT) has been widely adopted due to its ability to generate high-resolution cross-sectional images. The recent advancement of Photon-Counting Computed Tomography (PCCT) represents a major milestone in CT imaging, offering the ability to acquire energy-resolved data through photon-counting detectors (PCDs). This innovation enables improved material differentiation, enhanced contrast resolution, and reduced radiation dose. However, despite these advantages, PCCT faces several technical challenges such as charge sharing, pulse pile-up, spectral distortion, and noise amplification, which compromise image quality and quantitative accuracy, particularly in low-dose and sparse-view acquisitions. To address these limitations, this dissertation presents a series of deep learning and model-based approaches for PCCT data correction and image reconstruction. 

In Chapter one, the fundamentals of CT and PCCT imaging are introduced, along with key physical principles and reconstruction challenges. In Chapter two, a dual-domain correction framework is developed to jointly enhance data fidelity in both the projection and image domains. Specifically, a residual-based Wasserstein generative adversarial network (R-WGAN) is employed to suppress detector-related artifacts and photon-counting distortions in the projection domain, while an image-domain filtering strategy improves signal-to-noise ratio and texture preservation. In Chapter three, an adaptive tensor-based reconstruction approach is introduced, leveraging dynamic clustering and enhanced channel weighting for spectral-image similarity–based tensor decomposition with improved sparsity constraints. The proposed method adaptively prioritizes high-quality channels based on gradient magnitude, entropy, and spatial-spectral energy metrics, improving robustness under sparse-view and low-dose conditions. In Chapter four, a novel deep learning framework, termed Spectral Transformer Architecture with Bayesian Learning and Edge Preservation (STABLE-PCCT), is proposed to model spatial-spectral dependencies using depthwise separable convolutions, squeeze-and-excitation blocks, and a residual transformer module as a Bayesian prior. This architecture effectively reduces noise, maintains spectral consistency, and enhances structural details across energy channels. Extensive experiments on simulated and preclinical PCCT datasets demonstrate that the proposed methods significantly enhance image quality, spectral fidelity, and quantitative accuracy compared with state-of-the-art techniques. Collectively, this dissertation advances photon-counting CT by integrating physics-informed modeling, deep learning, and adaptive tensor reconstruction, contributing to high-fidelity, low-dose, and spectrally consistent imaging. Future research will focus on extending these frameworks to real-time reconstruction and multi-contrast imaging, and on validating the proposed methods in broader clinical settings to enable practical translation of photon-counting CT into routine healthcare.