Software Engineer Intern, Hyperfine Sequence Development Team

During the summer of 2021, I worked in Guilford, CT as a software development intern on the sequence development team at Hyperfine. I specifically worked on image reconstruction and analysis algorithms for Hyperfine's portable, low-field 64mT MRI imaging systems.

hyperfine.io

 Python, Matlab

 NumPy, SciPy, scikit-image, Matplotlib

Hyperfine Image

k-space Filtering

In lieu of large, conventional RF shielding rooms associated with typical diagnostic MRI machines, Hyperfine's low-field imager has a comparatively thin aluminum grid to shield signal from EM interference. To help correct for the resulting spike noise contributions from the surrounding hospital environment and other sources, I implemented and optimized a k-space filtering algorithm that removes signal contribution with frequencies that are substantially outside the Larmor frequency range.

Adaptive Coil Combination

Multiple coils work together in the imaging system for RF signals acquisition. The resultant complex datasets from each coil must be summed together to give the resulting image in position space. In practice, traditional coil combination algorithms involve estimating unknown coil sensitivity maps and often lead to phase discontinuities in the final combined image.

To overcome this limitation, I built on the previous work of Kellman and others to implement a method for adaptive coil combination that improves image SNR and ensures no phase singularities in the final image. Broadly, the algorithm is an iterative approach to singular value decomposition (SVD) of the local image space around any given pixel. I also further improved their proposed algorithm by implementing a mean filter with a spatially dynamic kernel to help further improve image SNR. Overall, my algorithm helped speed up the entire reconstruction pipeline by over 10% from existing methods, and improved image quality for both DWI and FLAIR images for over 60 clinical images collected.

Reconstruction Pipeline Tools

I wrote various scripts and data visualization tools to help speed up developer and research workflow.