11/05/2026
By Thomas Dugan

The Francis College of Engineering’s Department of Electrical and Computer Engineering invites you to attend a doctoral dissertation defense by Shuyi Fan, titled “Deep Learning for Medical Image Reconstruction and Diagnostic Analysis.”

Details

  • Date: November 23, 2026
  • Time: 10 a.m.
  • Location: Ball Hall, Room 302

Committee members

  • Hengyong Yu, Professor, Electrical and Computer Engineering, University of Massachusetts Lowell
  • Yan Luo, Professor, Electrical and Computer Engineering, University of Massachusetts Lowell
  • Orlando Arias, Assistant Professor, Electrical and Computer Engineering, University of Massachusetts Lowell
  • Rina Bao, Instructor, Department of Pediatrics, Boston Children’s Hospital and Harvard Medical School
  • Dayang Wang, Research Scientist, Subtle Medical Inc.

Abstract

Computed tomography (CT) images need to preserve anatomical details for reliable interpretation. This dissertation investigates deep learning-based approaches for limited-angle CT reconstruction, endoscopic polyp segmentation, and chest CT image analysis. It includes Multi-domain Mamba Network (MMN) for CT reconstruction, FOCUS-Med for endoscopic polyp segmentation, and a proposed study of AI-assisted chest CT interpretation.

MMN combines projection-domain refinement, structural image enhancement, multi-scale Mamba modeling, and projection-residual correction. Learned modules are connected by forward projection and filtered backprojection. The accompanying analysis explains selected operations through a reference energy.

Experiments are performed on XCAT phantom data, AAPM CT-MAR dataset, and a General Electric clinical dataset with patient-level separation to evaluate reconstruction accuracy, structural details, and computational cost. The results show that image error and structural similarity can favor different methods under severe angular incompleteness. MMN currently reconstructs two-dimensional slices. FOCUS-Med combines spatial and structural graph modeling, self-attention, and multi-scale feature fusion for endoscopic polyp segmentation.

The proposed study examines how AI can support recognition of findings in chest CT, particularly when findings are uncommon or occur together. It investigates how pretrained models can use information from relevant prior cases and evaluate performance on independent data. The reconstruction and analysis studies address related needs in medical imaging: preserving anatomical information and recognizing findings that inform diagnosis.