research

Real-time Magnetic Resonance Imaging

During my time at the University of Southern California, my research focused on developing new ways to improve real-time MRI technology. Specifically, I worked on making MRI scans faster, while still maintaining high image quality and accuracy.

3D real-time imaging

This project involved creating a real-time 3D MRI system that can capture moving vocal articulators (like the tongue and mouth) while someone is speaking. This breakthrough technique provides detailed insights into speech production, which can be used to improve language technology and treat speech disorders.

The below image demonstrates 3D tongue and vocal tract shape features that were never previously attempted to visualize.

Related publications

• Z Zhao, Y Lim, D Byrd, S Narayanan, and KS Nayak, Improved 3D real-time MRI of speech production, Magnetic Resonance in Medicine, vol. 85, no. 6, pp. 3182-3195, 2021.
• Y Lim, Y Zhu, SG Lingala, D Byrd, S Narayanan, and KS Nayak, 3D dynamic MRI of the vocal tract during natural speech, Magnetic Resonance in Medicine, vol. 81, no. 3, pp. 1511–1520, 2019.

Image deblurring

This project involved developing fast and efficient deep learning techniques to correct blurry images in real-time MRI scans. Our technique allows for high-quality, low-latency deblurring that is faster (<20msec per frame) and just as effective as traditional methods (>1sec).

Related publications

• Y Lim, S Narayanan, and KS Nayak, Attention-gated convolutional neural networks for off-resonance correction of spiral real-time MRI, in Proc. 28th ISMRM Scientific Sessions, Virtual Conference, April 2020.
• Y Lim, Y Bliesener, S Narayanan, and KS Nayak, Deblurring for spiral real-time MRI using convolutional neural networks, Magnetic Resonance in Medicine, vol. 84, no. 6, pp. 3438–3452, Dec. 2020.
• Y Lim, SG Lingala, S Narayanan, and KS Nayak, Dynamic off-resonance correction for spiral real-time MRI of speech, Magnetic Resonance in Medicine, vol. 81, no. 1, pp. 234–246, Jan. 2019.

A public speech production MRI dataset

Real-time MRI technology has greatly improved our understanding of speech science, linguistics, bioinspired speech technology, and clinical applications. However, access to this technology is limited, and more comprehensive datasets need to be made available.

As part of an interdisciplinary team, I helped develop a unique collection of multimodal speech production MRI data from a record-breaking 75 subjects. This dataset will be a valuable resource for researchers in the fields of linguistics, speech science, computational imaging, and engineering. It will provide them with the opportunity to gain deeper insights into human speech production and develop advanced computational imaging algorithms.

Related publications

• Y Lim, A Toutios, et al, A multispeaker dataset of raw and reconstructed speech production real-time MRI video and 3D volumetric images, Scientific Data (Nature) 8, 187. 2021.