Improving sub-pixel accuracy in ultrasound localization microscopy using supervised and self-supervised deep learning

With a spatial resolution of tens of microns, ultrasound localization microscopy (ULM) reconstructs microvascular structures and measures intravascular flows by tracking microbubbles (1-5 μm) in contrast enhanced ultrasound (CEUS) images. Since the size of CEUS bubble traces, e.g. 0.5-1 mm for ultrasound with a wavelength λ = 280 μm, is typically two orders of magnitude larger than the bubble diameter, accurately localizing microbubbles in noisy CEUS data is vital to the fidelity of the ULM results. In this paper, we introduce a residual learning based supervised super-resolution blind deconvolution network (SupBD-net), and a new loss function for a self-supervised blind deconvolution network (SelfBD-net), for detecting bubble centers at a spatial resolution finer than λ/10. Our ultimate purpose is to improve the ability to distinguish closely located microvessels and the accuracy of the velocity profile measurements in macrovessels. Using realistic synthetic data, the performance of these methods is calibrated and compared against several recently introduced deep learning and blind deconvolution techniques. For bubble detection, errors in bubble center location increase with the trace size, noise level, and bubble concentration. For all cases, SupBD-net yields the least error, keeping it below 0.1 λ. For unknown bubble trace morphology, where all the supervised learning methods fail, SelfBD-net can still maintain an error of less than 0.15 λ. SupBD-net also outperforms the other methods in separating closely located bubbles and parallel microvessels. In macrovessels, SupBD-net maintains the least errors in the vessel radius and velocity profile after introducing a procedure that corrects for terminated tracks caused by overlapping traces. Application of these methods is demonstrated by mapping the cerebral microvasculature of a neonatal pig, where neighboring microvessels separated by 0.15 λ can be readily distinguished by SupBD-net and SelfBD-net, but not by the other techniques. Hence, the newly proposed residual learning based methods improve the spatial resolution and accuracy of ULM in micro- and macro-vessels.

The role of innovative modeling and imaging techniques in improving outcomes in patients with LVAD

Heart failure remains a significant cause of mortality in the United States and around the world. While organ transplantation is acknowledged as the gold standard treatment for end stage heart failure, supply is limited, and many patients are treated with left ventricular assist devices (LVADs). LVADs extend and improve patients' lives, but they are not without their own complications, particularly the hemocompatibility related adverse events (HRAE) including stroke, bleeding and pump thrombosis. Mainstream imaging techniques currently in use to assess appropriate device function and troubleshoot complications, such as echocardiography and cardiac computed tomography, provide some insight but do not provide a holistic understanding of pump induced flow alterations that leads to HRAEs. In contrast, there are technologies restricted to the benchtop-such as computational fluid dynamics and mock circulatory loops paired with methods like particle image velocimetry-that can assess flow metrics but have not been optimized for clinical care. In this review, we outline the potential role and current limitations of converging available technologies to produce novel imaging techniques, and the potential utility in evaluating hemodynamic flow to determine whether LVAD patients may be at higher risk of HRAEs. This addition to diagnostic and monitoring capabilities could improve prevention and treatment of LVAD-induced complications in heart failure patients.

Cerebral microcirculation mapped by echo particle tracking velocimetry quantifies the intracranial pressure and detects ischemia

Affecting 1.1‰ of infants, hydrocephalus involves abnormal accumulation of cerebrospinal fluid, resulting in elevated intracranial pressure (ICP). It is the leading cause for brain surgery in newborns, often causing long-term neurologic disabilities or even death. Since conventional invasive ICP monitoring is risky, early neurosurgical interventions could benefit from noninvasive techniques. Here we use clinical contrast-enhanced ultrasound (CEUS) imaging and intravascular microbubble tracking algorithms to map the cerebral blood flow in hydrocephalic pediatric porcine models. Regional microvascular perfusions are quantified by the cerebral microcirculation (CMC) parameter, which accounts for the concentration of micro-vessels and flow velocity in them. Combining CMC with hemodynamic parameters yields functional relationships between cortical micro-perfusion and ICP, with correlation coefficients exceeding 0.85. For cerebral ischemia cases, the nondimensionalized cortical micro-perfusion decreases by an order of magnitude when ICP exceeds 50% of the MAP. These findings suggest that CEUS-based CMC measurement is a plausible noninvasive method for assessing the ICP and detecting ischemia.

Time-Resolved Echo-Particle Image/Tracking Velocimetry Measurement of Interactions Between Native Cardiac Output and Veno-Arterial ECMO Flows

Determination of optimal hemodynamic and pressure-volume loading conditions for patients undergoing veno-arterial extracorporeal membrane oxygenation (VA-ECMO) would benefit from understanding the impact of ECMO flow rates (QE) on the native cardiac output in the admixing zone, i.e., aortic root. This study characterizes the flow in the aortic root of a pig with severe myocardial ischemia using contrast-enhanced ultrasound particle image/tracking velocimetry (echo-PIV/PTV). New methods for data preprocessing are introduced, including autocontouring to remove surrounding tissues, followed by blind deconvolution to identify the centers of elongated bubble traces in images with low signal to noise ratio. Calibrations based on synthetic images show that this procedure increases the number of detected bubbles and reduces the error in their locations by 50%. Then, an optimized echo-PIV/PTV procedure, which integrates image enhancement with velocity measurements, is used for characterizing the time-resolved two-dimensional (2D) velocity distributions. Phase-averaged and instantaneous flow fields show that the ECMO flow rate influences the velocity and acceleration of the cardiac output during systole, and secondary flows during diastole. When QE is 3.0 L/min or higher, the cardiac ejection velocity, phase interval with open aortic valve, velocity-time integral (VTI), and mean arterial pressure (MAP) increase with decreasing QE, all indicating sufficient support. For lower QE, the MAP and VTI decrease as QE is reduced, and the deceleration during transition to diastole becomes milder. Hence, for this specific case, the optimal ECMO flow rate is 3.0 L/min.

A Novel Fabrication Method for Compliant Silicone Phantoms of Arterial Geometry for Use in Particle Image Velocimetry of Haemodynamics

Cardiovascular diseases (CVDs) are one of the leading causes of death globally. In-vitro measurement of blood flow in compliant arterial phantoms can provide better insight into haemodynamic states and therapeutic procedures. However, current fabrication techniques are not capable of producing thin-walled compliant phantoms of complex shapes. This study presents a new approach for the fabrication of compliant phantoms suitable for optical measurement. Two 1.5× scaled models of the ascending aorta, including the brachiocephalic artery (BCA), were fabricated from silicone elastomer Sylgard-184. The initial phantom used the existing state of the art lost core manufacturing technique with simple end supports, an acrylonitrile butadiene styrene (ABS) additive manufactured male mould and Ebalta-milled female mould. The second phantom was produced with the same method but used more rigid end supports and ABS male and female moulds. The wall thickness consistency and quality of resulting stereoscopic particle image velocimetry (SPIV) were used to verify the fidelity of the phantom for optical measurement and investigation of physiological flow fields. However, the initial phantom had a rough surface that obscured SPIV analysis and had a variable wall thickness (range = 0.815 mm). The second phantom provided clear particle images and had a less variable wall thickness (range = 0.317 mm). The manufacturing method developed is suitable for fast and cost-effective fabrication of different compliant arterial phantom geometries.

Image-Based Flow Simulations of Pre- and Post-left Atrial Appendage Closure in the Left Atrium

PURPOSE

For patients with atrial fibrillation, the left atrial appendage (LAA) is often the site of thrombus formation due to low atrial ejection fraction that triggers strokes and other thromboembolic events. Recently introduced percutaneous LAA occlusion procedure is known to reduce LAA-induced strokes. Despite having the procedure, there are still 11% of the patients who continue to suffer from future strokes or transient ischemic attacks, not accounting for the procedural related complications. The high failure rate is largely due to the variabilities in LAA's shape, size, and contractility which may result in ineffectiveness of this procedure. To correctly identify the candidates and evaluate the effectiveness of the procedure, we rely on patient-specific CT scans which provides the exact LA and LAA geometries and predictive hemodynamic analysis to assist in evaluating quantitative flow parameters pre- and post-LAA occlusion procedures. Hemodynamic parameters are critical to predict adverse hemodynamic flow patterns in LAA as well as the effectiveness of LAA closure in individual patient. The aim of this paper is to establish an image-based patient-specific computational fluid dynamic (CFD) simulation framework specific to the prediction of treatment outcomes of LAA closure with atrial fibrillation. This framework utilizes automated LA/LAA image segmentation which yields significant reduction in image processing. One set of patient data with successful procedure outcome is used to illustrate the potential of the proposed framework.

METHODS

The proposed LAA occlusion simulation framework is composed of several components: (1) a novel image segmentation procedure, which is fully-automated to identify LA/LAA geometries from CT images, (2) a finite-element mesh generation procedure which transforms the surface geometry into a 3-D volume mesh and properly identified boundary planes, (3) performing CFD simulations with atrial fibrillation flow boundary conditions, and (4) analyzing flow characteristics (velocity, flow patterns, streamlines, vortices) within the LA for before and after LAA closure.

RESULTS

Based on the LA/LAA segmentation of a 65 year old female patient with chronic atrial fibrillation, a CFD analysis was pursued to examine flow characteristics upon LAA closure. The results showed that the flow velocity magnitudes were significantly reduced by a maximum factor of 2.21, flow streamlines were greatly stabilized, and mitral outflow appeared to be more organized. Vortices were dramatically reduced in size, number, intensity, as well as duration. During diastole, the peak vortex diameter was reduced from 2.8 to 1.5 cm, while the vortex duration was reduced from 0.210 to 0.135 s. These flow characteristics all indicated a reduced risk in future thrombus formation and strokes based on the established relationship between flow and thrombus formation. For the patient case under study, the effectiveness of the procedure is predicted and found to be consistent with the actual procedural outcome.

CONCLUSIONS

This framework successfully predicted patient-specific outcome of a LAA closure procedure for one patient with atrial fibrillation. It can be further developed into a useful tool for pre-procedural planning and candidate selection. More patient data are necessary for further validation studies.

Editors' Choice Papers for 2018

No abstract.