Yaokun Shi with Advisor Prof. Kiruba S. Haran

Traditional MRI devices are considered one of the most powerful medical imaging tools that are non-invasive and offer robust information regarding soft tissues and pathological information. However, these devices are typically stationary and extremely costly, making them unsuitable for smaller clinics / point-of-care imaging. To address this issue, we propose a portable MRI system that features both hardware innovations and software pipelines to improve the image quality. This will significantly improve the cost of transportation, operation, and power consumption.

Three miniature Permanent Magnet Vernier Machines (PMVMs) are designed to rotate three magnet arrays at discrete steps, which serve as the base magnetic field as well as the spatial encoding scheme in replacement of the gradient coils. Such field patterns are simulated and confirmed through a 3D printing stage with custom software and a 3D hall-effect sensor. A solenoid transceiver radiofrequency (RF) coil is connected and driven by an FPGA-based microcontroller to excite the sample and receive information in the imaging field-of-view (FOV). Sweeping through a set of center frequencies and motor positions provides enough information about the imaging FOV and allows for a back-projection-based image reconstruction. The theoretical guarantees regarding the resolution and signal-to-noise ratio (SNR) are derived using the Projection-Slice theorem. Lastly, a self-supervised denoising and super-resolution machine learning framework is designed to enhance the resulting images to address the poor SNR and resolution issue, which makes the output images clinically useful.