In this paper we describe a pilot study, where machine learning methods are used to differentiate between congenital heart diseases. Our approach was to apply convolutional neural networks (CNNs) to echocardiographic images from five different pediatric populations: normal, coarctation of the aorta (CoA), hypoplastic left heart syndrome (HLHS), transposition of the great arteries (TGA), and single ventricle (SV). We used a single network topology that was trained in a pairwise fashion in order to evaluate the potential to differentiate between patient populations. In total we used 59,151 echo frames drawn from 1,666 clinical sequences. Approximately 80% of the data was used for training, and the remainder for validation. Data was split at sequence boundaries to avoid having related images in the training and validation sets. While training was done with echo images/frames, evaluation was performed for both single frame discrimination as well as sequence discrimination (by majority voting). In total 10 networks were generated and evaluated. Unlike other domains where this network topology has been used, in ultrasound there is low visual variation between classes. This work shows the potential for CNNs to be applied to this low-variation domain of medical imaging for disease discrimination.
Coarctation of aorta (CoA) is a critical congenital heart defect (CCHD) that requires accurate and immediate diagnosis and treatment. Current newborn screening methods to detect CoA lack both in sensitivity and specificity, and when suspected in a newborn, it must be confirmed using specialized imaging and expert diagnosis, both of which are usually unavailable at tertiary birthing centers. We explore the feasibility of applying machine learning methods to reliably determine the presence of this difficult-to-diagnose cardiac abnormality from ultrasound image data. We propose a framework that uses deep learning-based machine learning methods for fully automated detection of CoA from two-dimensional ultrasound clinical data acquired in the parasternal long axis view, the apical four chamber view, and the suprasternal notch view. On a validation set consisting of 26 CoA and 64 normal patients our algorithm achieved a total error rate of 12.9% (11.5% false-negative error and 13.6% false-positive error) when combining decisions of classifiers over three standard echocardiographic view planes. This compares favorably with published results that combine clinical assessments with pulse oximetry to detect CoA (71% sensitivity).
Two-dimensional color Doppler echocardiography is widely used for assessing blood flow inside the heart and blood vessels. Currently, frame acquisition time for this method varies from tens to hundreds of milliseconds, depending on Doppler sector parameters. This leads to low frame rates of resulting video sequences equal to tens of Hz, which is insufficient for some diagnostic purposes, especially in pediatrics. In this paper, we present a new approach for reconstruction of 2D color Doppler cardiac images, which results in the frame rate being increased to hundreds of Hz. This approach relies on a modified method of frame reordering originally applied to real-time 3D echocardiography. There are no previous publications describing application of this method to 2D Color Doppler data. The approach has been tested on several in-vivo cardiac 2D color Doppler datasets with approximate duration of 30 sec and native frame rate of 15 Hz. The resulting image sequences had equivalent frame rates to 500Hz.
While Moore's law has eliminated the need for algorithm optimization when computing 2D dynamic contours, real-time
3D image analysis remains limited by computational bottlenecks. We are specifically concerned with segmenting
3D volumetric ultrasound streams from echo-cardiograph machines (Phillips Medical Systems, Andover, MA) for
analysis of cardiac function. The system uses a 3000 element array that produces 20-25 volumes per second at
a resolution of 128x48x204 voxels. This yields a data rate of of 240 Mbits/sec, requiring efficient algorithms and
implementations to track moving cardiac tissue in real-time.
This paper discusses implementation of active 2D deformable models for real-time volumetric segmentation at
the high data rates described above. We demonstrate that using an efficient approximation of local curvature change
in the implementation of dynamic contours leads to real-time volumetric segmentation on mid-range off-the-shelf
hardware without the use of specialized graphics hardware.
Our dynamic contour implementation relies on an optimal estimation of local curvature change based on a Menger
curvature calculation. We investigate the role of curvature approximations and smoothness with respect to optimal
contour point motions and step size in real-time implementations. This smoothness provides reasonable shape estimates
in the absence of appropriate or conflicting external image input.
Finally, we present a 3D image segmentation algorithm based on an efficient implementation of 2D dynamic
contours, and demonstrate real-time performance with high volumetric data rates.
Surgical repair of the mitral valve is preferred in most cases over valve replacement, but replacement is often performed
instead due to the technical difficulty of repair. A surgical planning system based on patient-specific medical images that
allows surgeons to simulate and compare potential repair strategies could greatly improve surgical outcomes. In such a
surgical simulator, the mathematical model of mechanics used to close the valve must be able to compute the closed state
quickly and to handle the complex boundary conditions imposed by the chords that tether the valve leaflets. We have
developed a system for generating a triangulated mesh of the valve surface from volumetric image data of the opened
valve. We then compute the closed position of the mesh using a mass-spring model of dynamics. The triangulated mesh
is produced by fitting an isosurface to the volumetric image data, and boundary conditions, including the valve annulus
and chord endpoints, are identified in the image data using a graphical user interface. In the mass-spring model, triangle
sides are treated as linear springs, and sides shared by two triangles are treated as bending springs. Chords are treated as
nonlinear springs, and self-collisions are detected and resolved. Equations of motion are solved using implicit numerical
integration. Accuracy was assessed by comparison of model results with an image of the same valve taken in the closed
state. The model exhibited rapid valve closure and was able to reproduce important features of the closed valve.
Parametric active deformable models for image-based segmentation offer a distinct advantage over level sets: speed. This paper presents an extension to active deformable models that makes real-time volume segmentation possible on mid-range off-the-shelf hardware and without the use of specialized graphics hardware. The proposed method uses region-based parametric deformable models. A region-based parametric model, represented by a polygon, must remain non-self intersecting (simple) while undergoing deformation. The simplicity constraint can be enforced by allowing topological changes or by restricting motions of the curve. In either case, intersections of curve segments must be detected otherwise catastrophic divergence results. Good performance relies on the efficiency of the intersection check operation. This paper presents a parameter-free and efficient technique for on-line simplicity checking of polygons undergoing motion. We present timing results validating our approach; in particular, we segment 3-D ultrasound data at 20 volumes per second.