Handheld MRI with Rotational Spatial Encoding and Deep-Learning Priors
MS student Yaokun Shi with Advisor K. Haran
Funded by the Grainger CEME, our hand-held MRI replaces a conventional C-shaped or solenoid magnet with two diametrically magnetized cylindrical magnets capable of individually rotating. This unique spatial encoding creates a non-homogeneous magnetic field in the imaging volume and thus compensates for the removal of the gradient coils. In traditional MRI devices, these coils contribute to the power and size cost. As
11 shows, the current Tx and Rx coils, along with all the other control components, are embedded in the custom-printed circuit board, which then communicates with the microcontroller boards to send signals serially to the host PC.
The next step in the project involves precise measurements of the rotating magnetic field, as demonstrated in Figure 12. The coils will be tested using Larmor frequencies generated by this magnetic field on phantoms filled with doped water. PETRA (pointwise encoding time reduction with radial acquisition), a unique scanning sequence used in traditional MRI devices, is the protocol to acquire data and process coil signals. It requires no gradient coils and produces an image with T1 contrast. This protocol gives a low-quality image (low SNR) due to the low magnetic field, so the plan is to implement deep-learning methods to enhance the image quality. As the neural network requires heavy computing power, and we do not wish to add extra components or computing units to the device, we intend to generate these intelligible images through cloud-computing platforms.