MRI scanners are largely categorized by their magnetic field strengths, which determine the scans' resolution and clarity along with the cost, size, and weight of the machine. The most common MRIs seen around hospitals in the US are 1.5 or 3 Tesla models which range from $1 to $2.5 million and require an exceeding amount of power and space. In hopes of increasing MRI accessibility, recently, we've seen the development and approval of low-field scanners, which although much lower in price and allow for portability, also decreases the resolution and increases the noise in the scans.

This project aims to better understand the capabilities of these lower-resolution images by creating a pipeline to simulate samples of low-field imaging. While low-field scanners are exponentially lower in cost, size, and weight, due to its novelty, as of right now, there are only five centers with access to this 64 mT low-field scanner. Fortunately, Penn is among one of those five and has the ability to analyze and understand the potential of this portable scanner. Even so, the low-field dataset size pales in comparison to the plethora of 3T scans that have been collected throughout the years, spanning several patient populations. Thus, this application of GANs takes advantage of the vast 3T dataset to ideally create simulated trials so we can evaluate what different patient populations' scans may look like on a low-field scanner.