Multimodal fusion of US and MR images for Twin to Twin Transfusion Syndrome fetal surgeries

Abstract

Around 15% of all monochorionic pregnancies are affected by Twin to Twin Transfusion Syndrome. It involves chronic transfusion between donor twin to recipient twin, leading to an unbalanced growth of fetus due to unequal sharing of the placental blood volume. There are also other associated issues. The donor is at high risk of kidney problems related to not having enough blood flow. The extra blood flow that the recipient twin receives can put a strain on the heart leading to heart failure or developmental problems. Both twins can also have neurodevelopmental difficulties due to inadequate oxygen and nutrients as a result of blood transfusion. For untreated cases, the mortality of twins increases by around 90%. In treated cases, intrauterine surgery tries to stop blood transfusion by which some vascular connections are coagulated. Before the intervention takes place, surgeons have to plan it to ensure the correct localization of the placental vessel. Magnetic resonance (MR) is used to visualize the intrauterine structures, but the Ultrasound (US) technique is used to localize vessels because it gives accurate knowledge about vascularization. The aim of this study is to develop a multimodal registration algorithm to combine MR and US information. Three different registration strategies were developed, so placentas for both modalities can be aligned automatically. To do so, free deformable and rigid transformations were used in combination with different input data, such as medical images, segmentation or image distance maps. Then, algorithms were tested on ten different patients from Hospital de la Maternitat, and results were visually and quantitatively assessed. For visual evaluation, MITK software is used to compare the overlapping between US and MR placentas, before and after registration. Algorithms are also verified by some quantitative metrics (Dice, Volume Similarity, Hausdorff distance, and umbilical cord insertion distances) that measure the overlapping and surface distances between registered placentas. Some limitations regarding provided images and data have been taking into account after registration. Results obtained suggest that a high level of deformation is produced when free deformable strategies and segmentations were used. As a result, vessels morphology can be lost in some cases. When greater preservation of the shape is needed distance map images in combination with rigid strategies should be implemented. Beside intrinsic variations, developed strategies have shown considerable advances in the registration process, thus being potential tools for improving the pre-clinical planning of the TTTS intervention.