ساراسادات موسويان

Dynamic magnetic resonance imaging (MRI) plays a vital role in the diagnosis an‎d study of physiological processes in the human body. However, the inherent challenges in acquiring high-quality data within a short time highlight the need for advanced image reconstruction techniques. This imaging modality often encounters difficulties such as noise, motion artifacts, an‎d long scan times, which make high-quality reconstruction challenging. Furthermore, the high cost of imaging an‎d patient discomfort have led to an increasing interest in recent years in fast an‎d undersampled reconstruction methods. In this study, the fundamental concepts of dynamic MRI, its major challenges, an‎d conventional reconstruction approaches are first reviewed. A novel low-rank based method is then proposed, which combines the structural properties of the data with rank reduction an‎d sparsity principles to reconstruct high-quality images from undersampled data. To enhance stability an‎d suppress noise, the proposed algorithm employs adaptive regularization strategies along with complementary denoising techniques. In this framework, the reconstruction problem is formulated as a cost function with two main terms: a low-rank component for modeling temporal correlations an‎d a sparsity component for capturing spatial features. The optimization is solved using half-quadratic splitting combined with adaptive thresholding, while Nesterov acceleration an‎d multi-stage denoising are incorporated to further improve efficiency. Moreover, the convergence of the algorithm an‎d its robustness to noise are analytically investigated an‎d validated. Experimental results on both real an‎d simulated datasets demonstrate that the proposed method improves spatial an‎d temporal resolution, exhibits robustness against noise an‎d motion variations, an‎d achieves competitive performance compared to reference approaches. In addition to image quality enhancement, reduced reconstruction time an‎d the potential for clinical applicability are key outcomes of this work. Overall, the findings indicate that employing a low-rank model-based approach can serve as an effective step toward improving the quality of medical imaging, accelerating reconstruction, lowering costs, an‎d enhancing diagnostic processes in biomedical engineering.

تصويربرداري تشديد مغناطيسي , بازسازي تصوير , يادگيري عميق , حسگري فشرده , مدل‌هاي رتبه پايين

Magnetic Resonance Imaging , Image Reconstruction , Deep Learning , Compressed Sensing , Low-Rank Models

Sara Sadat Mousavian

Aboozar Ghaffari