Towards an Automatic Super-Resolution Image Based Method for Current 2-Dimensional Contrast Enhanced Ultrasound

Abstract

This study aims to create a new automatic super-resolution image based method to be used with current contrast enhanced ultrasound imaging (CEUS) technology without the need for the generation of new imaging modes. The emerging field of super-resolution ultrasound imaging aims to detect and track single microbubbles to provide image data that reveals the detail of the vascular treewith micrometre resolution. Current 2-dimensional (2D) ultrasound provides significant challenges as the point spread function (PSF) and the number of MBs are highly variable not only between different image frames but also throughout a single image. This high variability is related to the orders of magnitude difference in blood volume between the largest (arteries) and the smallest of vessels (capillaries). The original hypothesis of this study was that a super-resolution image based method can provide up to 10-fold resolution gains compared to conventional imaging using video loops that are short in duration and are generated with existing CEUS modes. For this a new super-resolution image based method was developed. The method is based on a morphological based algorithm designed forfluorescence optical microscopy and incorporates both particle detection and trajectory linking and is also capable of handling large particle numbers (Wilson et al. Royal Soc Open Sci 2016; 3: 160225). The study was designed to address the original hypothesis using an in vivo animal model. The ovaries from blackface ewes were exteriorised and the corpus luteum (CL) was imaged using CEUS. The CL is ideally suited for perfusion imaging studies as it is a highly angiogenic gland and has the size of a small solid tumour (Sboros et al. Ultrasound Med Biol 2011; 37: 59). Once the CEUS imaging was complete the CL was fixed for histological evaluation in order to identify large vessels and endothelium for comparison with the CEUS image data. An iU-22 Philips scanner with an L9-3 linear array probe (6MHz) was used to generate the CEUS images. Intravenous infusions of Sonovue MB suspensions generated images with high densities of single microbubbles. Video loops (typically 1 minute long) were collected. The image analysis method for generating the super-resolution data utilises 11 different parameters. Firstly, the method detects and localises microbubbles through the use of a particle probability image (PPI) which enhances their signal. The PPI is formed using non-local and statistical mapping of the original greyscale by measuring local contrast and particle size. In the second step microbubbles in consecutive frames are linked in order to generate their trajectories through the vascular bed. The nearest neighbour linking model used has been extended to include the morphology and intensity profiles of the microbubbles. The software is automatic in that after an initial manual training of the parameter set it is able to adapt to the image data. Several less than 1 minute long (less than 600 frames) video loops of CEUS image data from the CL were processed. The processing of each video loop using the method described above produces a particle/microbubble density and a velocity map. These were compared with the histological evaluation that determines the location of the lectin stained capillaries and the actin stained arteries. The processed images provided a 5 to 10-fold improvement in resolution when compared to conventional ultrasound. Entire arteries with diameters as small as 200μm were visualised with a less than 20% error in their measure diameter, compared to the spatial resolution of conventional ultrasound which is only 400μm in optimal conditions. In addition, small arterioles in the range of 40μm can also be visualised. Importantly, the super-resolution ultrasound ethod handled on average over 40,000 single scatter events per video loop. In individual image frames it was possible to identify up to 300 single microbubbles in the 1.5 mm x 1.5 mm region of interest of the CL. The new image analysis method enables super-resolution maps of the vascular bed using short duration video loop data captured by conventional 2D CEUS, confirming our original hypothesis. The robust detection and linking of microbubbles enables the automatic handling of large microbubblenumbers in each image, which in turn significantly reduces the required duration of the examination. This method is ready to be tested in patient data.