Presenter:
Samira Vafay Eslahi is a Ph.D. candidate at the Department of Electrical and Computer Engineering at Texas A&M University. She successfully defended her dissertation and is expecting to graduate in August 2022. Her research focuses on developing artificial intelligence algorithms for medical image computation and analysis. In her Ph.D. studies, she proposed a novel evolutionary-based Neural Architecture Search strategy to optimize deep learning models for MRI scan acceleration.
Time:
2:00 PM – 3:00 PM CDT, July 27th
Description:
Magnetic Resonance Imaging (MRI) is a premier non-ionizing medical imaging technique that can produce high-quality images. However, the main drawback of MRI is its long scanning procedure compared to other modalities, such as ultrasound and computed tomography, resulting in high motion artifact and patient discomfort. Considering the physical and physiological limitations of the MRI machines, undersampling the data, which ultimately will be reconstructed to the image of the fully sampled data, is one of the most popular accelerating methods. Among all the reconstruction algorithms introduced in the literature, deep learning algorithms showed promising results in producing high-quality images, leading to a shorter scan and reconstruction time. While deep learning is a powerful method in image reconstruction, the network’s performance significantly relies on the architecture of the network. However, finding a network that accurately reconstructs an image that resembles the reference is challenging due to a large number of possible architectures. A neural architecture search algorithm is able to automate the network’s architecture design by efficiently searching for the best reconstruction model in the search space. This presentation discusses neural architecture search applications in MRI reconstructions to generate high-quality images through automatically designing deep learning models.