Doing a biopsy or injection on a small target in the liver is challenging, because of the movement in and around the liver due to the respiratory motion (RM). This small target requires an imaging modality with a high contrast, like MR, to accurately locate the target. At the mean time, MRI has a large acquisition time (1 per second) and cannot be used for real-time guidance.
A solution to this problem is to let the patient hold their breath. The patient must be able to hold their breath for around twenty seconds, which can be uncomfortable for the patient and some might be unable to do so. This also drastically increases surgery time and the location of the target might be inconsistent between breath holds.
Another solution is to compensate for the motion by using a surrogate signal. This signal can acquire data with a low acquisition time to model the respiratory motion, and use learning algorithms to correlate the motion captured by the surrogate signal, with the target's location, captured by an MR scan.
Some surrogate signals that have previously been used for this goal are: a reference needle, ultrasound images, spirometers, respiratory bellows and optical tracking of markers on the skin.
Most of the solutions returned an x, y, and z coordinate that can be used by, for example, a robot to accurately locate the target. Another solution used Kernal Density Estimation and ultrasound images as a surrogate signal, to create synthetic MR scans. These scans can assist a surgeon by giving real-time guidance.
The goal of this project is to also create synthetic MR scans by using a state-of-the-art generative adversarial network (GAN) together with a surrogate signal.