I am a Ph.D. student at Stanford University where I work on computational magnetic resonance imaging and artificial intelligence. I am advised by Shreyas Vasanawala and funded by the NSF Graduate Research Fellowship.
I did my undergrad at University of Southern California where I worked with Krishna Nayak on cardiac MRI for a few years. I've also interned at the National Institutes of Health and GE Healthcare. This summer I will be working at Apple AI Research
News
January 17, 2020: Our tutorial on deep learning-based MR image reconstruction was published in IEEE Signal Processing Magazine! See paper and code.
Research
Broadly, I'm interested in developing signal processing and artificial intelligence approaches to solve problems in computational sensing. Currently, I focus on upstream AI methods to improve pediatric magnetic resonance imaging with respect to speed and diagnostic accuracy. Representative papers are highlighted.
Accelerating dynamic MRI using a physics-driven deep learning reconstruction Christopher Sandino,
Peng Lai,
Shreyas Vasanawala,
Joseph Cheng
Presented at ISMRM 2019 (Magna Cum Laude). Submitted to MRM, 2019.
Designed a convolutional neural network architecture to reconstruct dynamic MR images from highly undersampled raw data by jointly leveraging physics-based signal models and deep 3D priors. Using this technique, standard MRI scans which once took 5-6 breath-holds now take a single breath-hold allowing for faster and higher accuracy assessment of heart function.
Developed a motion-robust MRI acquisition scheme to enable measurement of velocity vector fields as a function of 3D space and time. We leverage low-rank compressed representations and a stochastic algorithm to reconstruct 100GB datasets acquired from each scan. Our technique has enabled fast, free-breathing, and comprehensive abdominal MRI exams for pediatric patients in as little as 5 minutes.
Cardiac tissue characterization using magnetic tissue parameter mapping Christopher Sandino,
Peter Kellman,
Andrew Arai,
Michael Hansen,
Hui Xue
Presented at ISMRM 2015. Published in JCMR, 2015.
Developed a more robust method for estimating T2*, a magnetic tissue parameter which can be used to identify iron overload in the heart. We also formulated a method to derive "confidence" maps using noise statistics measured with a short and simple calibration scan. This technique is implemented in medical image reconstruction framework, Gadgetron, and is being used at over 30 clinical sites worldwide.
