We propose a deep learning approach for the visualization of vessels during image-guided procedures without the need for contrast agents or pre-operative 4D CTs.
X-ray images are routinely acquired during image-guided surgical interventions, but the vessels are not visible in these images. Injection of contrast agents is traditionally used to visualize the vessels, but only a limited dose may be injected before it is toxic to the patient. Another solution is to project the segmented vessels from the pre-operative CT-scan. However, the anatomy of the patient during the intervention is different from the anatomy in the pre-operative CT-scan.
Using a neural network, we can predict the deformation of the anatomy in the X-ray image to superpose the vessels on the image at the right place and in the right shape. The hepatic vein position (in red in the video) is not corrected to follow the breathing motion, and is far from the true position. The hepatic vain veins (in green in the video) are deformed using our neural network prediction and follow the breathing motion visible in the image.
