Detection of motion during tomographic acquisition by an optical flow algorithm.

An automatic procedure to detect and quantify patient motion during the acquisition of the tomographic views in single photon emission computerized tomography (SPECT) is proposed. This method first computes the optical flow vector field which assigns to each pixel of a tomographic view the two-dimensional displacement vector that describes its motion between two successive views. The average optical flow in a region of interest is then computed to measure its inter-view global motion. This algorithm is tested on a point source, on a cardiac phantom (with and without induced motion), and on a patient. The proposed method can accurately detect the presence of motion, localize the camera angle at which motion occurred, and measure the distance of motion. The optical flow method can be used to control the quality of the tomographic acquisition and to alert the user to the potential of reconstruction artifacts due to patient motion. It can also be used to correct for the translational motion in the direction of the axis of rotation of the camera.

An expectation maximization reconstruction algorithm for emission tomography with non-uniform entropy prior.

A Bayesian image reconstruction algorithm is proposed for emission tomography. It incorporates the Poisson nature of the noise in the projection data and uses a non-uniform entropy as an a priori probability distribution of the image in a maximum a posteriori (MAP) approach. The expectation maximization (EM) method was applied to find the MAP estimator. The Newton-Raphson numerical method whose convergence and positive solutions are proven, was used to solve the EM problem. The prior mean at iteration k was determined by smoothing the image obtained at iteration k-1. Comparisons between the ML and the MAP algorithm were carried out with a numerical phantom that contains a narrow valley region. The ML solution after 50 iterations was chosen as the initial solution for the MAP algorithm, since the global performance of the ML algorithm deteriorates with increasing number of iterations while its local performance in the valley region is always improving. The resulting algorithm is a compromise between ML who has the best local performance in the valley region and the MAP who has the best global performance.

Restoration of the velocity field of the heart from two-dimensional echocardiograms.

A method to quantify the motion of the heart from digitized sequences of two-dimensional echocardiograms (2-D) echos was recently proposed. This method computes on every point of the 2-D echoes, the 2-D apparent velocity vector (or optical flow) which characterizes its interframe motion. However, further analysis is required to determine what part of this motion is due to translation, rotation, contraction, and deformation of the myocardium. A method to locally obtain this information is presented. The proposed method assumes that the interframe velocity field U(xy), V(x,y) can be locally described by linear equations in the form U(x,y)=a+Ax+By; V(x,y)=b+Cx+Dy. The additional constraint was introduced in the computation of the local velocity field by the method of projections onto convex sets. Since this constraint is only valid locally, the myocardium must be first divided into sectors and the velocity fields computed independently for each sector.

Local histogram information content of ultrasound B-mode echographic texture.

In the diagnostic ultrasound community, the echographic B-scan texture is an important area of investigation since it can be analyzed to characterize the histologic state of internal tissues. In the present paper, a minicomputer based system was used to digitize B-mode images and to develop a method to measure their textural information. This method is based on the concept of local information content of spatial image proposed by Lowitz (1983, 1984). It first measures the local gray-level histogram in a small square window centered on each picture element (pixel) of a digitized B-mode image. The information derived from the local histograms is then used to characterize the tissues, to partition the B-mode image into homogeneous zones of texture, to estimate to what extent a tissue is different from another, to delimit the contours of a tissue and to measure its surface. The method is illustrated on the thyroid gland but it can be applied to the study of other organs.

Texture analysis of ultrasound B-mode images by segmentation.

B-mode texture characterization by supervised methods of pattern recognition is subject to the following drawbacks: precise localization of the lesion to characterize is often difficult and, even when the lesion is well isolated, its texture can be corrupted by the presence of tumor non specific structures. These structures are not easily discernable and introduce a bias in the statistical measures. The results presented in this paper show that these problems can be circumvented by the use of an unsupervised method of image segmentation. The method enhanced the B-mode image and partitions the non specific structures and the lesion texture in different regions which can be characterized independently by statistical methods. The unsupervised approach also facilitates the clinical diagnosis done by visual inspection, by revealing subtle characteristics of the B-mode textures.

A simulation study of the effects of torso inhomogeneities on electrocardiographic potentials, using realistic heart and torso models.

The effects of torso inhomogeneities on electrocardiographic potentials were investigated via computer stimulation, using a 23-dipole heart model placed within a realistically shaped human torso model. The transfer coefficients relating the individual dipoles to the torso surface potentials, as well as the body surface potential maps, the vectorcardiogram, and the 12-lead electrocardiogram resulting due to normal activation of the heart model, were calculated for each of the following torso conditions: homogeneous, homogeneous + skeletal muscle layer, homogeneous + muscle layer + lungs, and homogeneous + muscle layer + lungs + intraventricular blood masses. The effects of each inhomogeneity were deduced by comparing results before and after its inclusion. For individual dipole transfer coefficients we confirm the validity of the "Brody effect," whereby the high conductivity blood masses augment radially oriented dipoles and diminish tangentially oriented ones. With regard to the vectorcardiogram , the electrocardiogram, and the body surface potential maps, the major qualitative effects were an augmentation of the head-to-foot component of the vectorcardiogram due to the lungs, and a smoothening of notches in the electrocardiogram (temporal filtering) and of isopotential contours in the body surface potential maps (spatial filtering) with a consequent loss of information, due to the blood masses, muscle layer, and, to a lesser extent, the lungs. Besides the above qualitative effects of the inhomogeneities, there were also large quantitative effects on the surface potentials, namely, magnitude increases due to the blood masses and magnitude decreases due to the muscle layer, that--if unaccounted for--could compromise the inverse solution of these potentials for the cardiac dipole sources.

A comparative evaluation of three different approaches for detecting body surface isopotential map abnormalities in patients with myocardial infarction.

Three approaches for detecting abnormalities in body surface potential maps recorded from patients with myocardial infarction were evaluated. The maps are generated from 26 simultaneously recorded unipolar electrocardiograms. All three approaches detect the deviations in certain parameters from control values determined from 50 normal subjects. The first approach emphasizes qualitative deviations in the trajectories of the surface potential map extrema during QRS and correctly classified all but one infarct in a test group comprising 30 normals and 30 cases of myocardial infarction. The second approach classifies a test subject as abnormal if any one of his 26 lead waveforms deviates appreciably at any instant during QRS from the mean waveform for the particular lead plus or minus two standard deviations, these being determined from the control group. This method, while correctly identifying all infarcts, resulted in a large number of false positives, misclassifying 22 of 30 normals. A final method was to obtain an instant by instant plot of the correlation coefficient between the mean surface potential map during QRS for the 50 normals and that of the subject being tested. Test cases were classified as abnormal if any correlation coefficient value fell below an envelope determined from the correlation coefficient plots obtained by correlating the maps of all 50 normals with their own mean. Twenty-nine normals and 26 infarcts were correctly classified. On the basis of these results, the first approach is superior to the other two for detecting surface potential map abnormalities in patients with myocardial infarction.