BME PhD Preliminary Exam Announcement for Farhan Sadik (R. Surowiec, advisor)

Everyone is invited to attend the public presentation beginning at 11:00 AM.

Title: Motion Correction in High-Resolution Medical Imaging: From Modality-Specific Solutions in HR-pQCT and UTE-MRI Toward Generalized Medical Image Reconstruction

Date: December 15^th, 2025

Time: 11:00 AM

Location: MJIS 1083

Microsoft Teams Link: Farhan's Tentative Preliminary Exam | Meeting-Join | Microsoft Teams

Committee Members: Dr. Rachel Surowiec (Chair), Dr. Joseph Wallace, Dr. Young Kim, Dr. Fengqing Zhu, and Dr. Uzay Emir

Abstract: Despite advances in foundation-model–driven medical image segmentation, classification, and synthesis, motion remains a major bottleneck, where even minor patient movement can render scans unusable. Current solutions, such as rescanning or discarding corrupted scans, are costly and inefficient. This proposal focuses on developing an automated framework for motion correction and image restoration across imaging modalities, beginning with two specific modalities: high-resolution peripheral quantitative computed tomography (HR-pQCT) for bone imaging and ultrashort-echo time magnetic resonance imaging (UTE-MRI) for pulmonary imaging. For HR-pQCT, I construct a physics-driven motion simulation pipeline that perturbs motion-free scans according to the CT data-acquisition physics, enabling the generation of a large, diverse dataset with controllable motion artifacts. Building on this dataset, I introduce an edge-enhanced self-attention Wasserstein generative adversarial network (ESWGAN) to correct rigid motion in the simulated domain. Furthermore, to transfer the learned motion correction capability to real-world scans, I introduce a Dual-CycleGAN architecture that adversarially narrows the gap between simulated and real corrupted images, thereby enabling robust correction of motion artifacts in real-world HR-pQCT data. For non-rigid motion in free-breathing neonatal UTE-MRI, I utilize a novel petal-like rosette trajectory (PETALUTE), combined with k-space–based motion signal extraction and iterative compressed sensing to achieve motion-compensated reconstruction. Finally, I aim to explore generalized MRI/CT reconstruction under varying degradations, moving toward foundational models for inverse imaging problems. Preliminary results demonstrate state-of-the-art performance: in HR-pQCT, motion correction achieves 5.42% improvement in PSNR from the baseline, while the domain adaptation framework successfully transfers these performances to real-world scans. In neonatal pulmonary imaging, PETALUTE acquisition yields superior motion-compensated reconstruction compared to traditional radial UTE acquisition. Overall, my work addresses motion-induced image degradation, preserves valuable datasets, cuts clinical costs, and has the potential to establish a foundation for generalized medical image reconstruction.