Nina Montana-Brown - Mark Pinnock - CMIC/WEISS joint seminar series

Mark Pinnock

Title: Combined 3D super-resolution, de-noising and partial volume correction for percutaneous ablation

Abstract: Interventional computed tomography (iCT) is a commonly used imaging modality for renal cryoablation, in part due to its excellent anatomical visualisation3 as well as its competitive cost and availability in comparison to magnetic resonance (MR) imaging. However, no imaging system is free from artefacts or resolution constraints. To reduce the radiation exposure during iCT-guided procedures, the X-ray dose is decreased and the slice thickness is increased, resulting in increased noise and decreased spatial resolution in the z-direction. Super-resolution (SR) is the application of post-processing techniques in an attempt to improve the above drawbacks in reducing radiation dose. The aim of this work is to take the thick slice, noisy, low quality (LQ) images and convert them to thin slice, high quality (HQ) images. To achieve this, we make the following contributions: 1) We modify the 3D U-Net with an up-sampling module that allows inputs and outputs of differing dimensions and trains on the entire image volume rather than patches; 2) As well as up-sampling in the z-direction, our proposed method also performs de-noising in the xy-plane; 3) Finding matched LQ-HQ pairs within iCT data is challenging owing to tissue deformation, respiratory motion, field of view (FOV) changes and movement within the scanner. We therefore train on simulated low quality (sLQ) data and show that this enables us to generalise to performing SR on real low quality (rLQ) images, tested on the scarce rLQ-HQ pairs.

Nina Montana-Brown

Title: Vessel Segmentation for Automatic Registration of Untracked Laparoscopic Ultrasound to CT of the Liver

Abstract: Registration of Laparoscopic Ultrasound (LUS) to a pre- operative scan such as Computed Tomography (CT) using blood vessel information has been proposed as a method to enable image-guidance for laparo- scopic liver resection. Currently, there are solutions for this problem that can potentially enable clinical translation by bypassing the need for a manual initialisation and tracking information. However, no reliable framework for the segmentation of vessels in 2D untracked LUS images has been presented.

We propose the use of 2D UNet for the segmentation of liver vessels in 2D LUS images. We integrate these results in a previously developed registration method, and show the feasibility of a fully automatic untracked LUS to CT registration that does not require an initialisation.

We validate our segmentation using LUS data from 6 patients. We test multiple models by placing patient datasets into different combinations of training, testing and hold-out, and obtain mean Dice scores ranging from 0.543 to 0.706. Using these segmentations, we obtain registration accuracies between 6.3 and 16.6 mm in 50% of cases.