Christopher D. Garson

Ultrasound Despeckling for Contrast Enhancement

Images produced by ultrasound systems are adversely hampered by a stochastic process known as speckle. A despeckling method based upon removing outlier is proposed. The method is developed to contrast enhance B-mode ultrasound images. The contrast enhancement is with respect to decreasing pixel variations in homogeneous regions while maintaining or improving differences in mean values of distinct regions. A comparison of the proposed despeckling filter is compared with the other well known despeckling filters. The evaluations of despeckling performance are based upon improvements to contrast enhancement, structural similarity, and segmentation results on a Field II simulated image and actual B-mode cardiac ultrasound images captured in vivo.

3D prostate elastography: algorithm, simulations and experiments.

A multi-resolution hybrid strain estimator is presented. The estimator is locally initialized by the B-mode tracking stage. Nonlinear and linear stretching regimes are applied in successive RF tracking stages for refining the estimated axial and lateral displacements. A staggering operator is used to derive the strain images from the reconstructed axial displacements. Simulations and experiments, conducted at a center frequency of 12 MHz, 40% fractional bandwidth, on a 128 element transducer with 0.2 mm pitch, with elastographic window length of 2 mm and overlap of 90%, demonstrate a 3-6 dB improvement in the elastographic contrast-to-noise ratio over the results obtained using conventional multi-stage stretching based strain estimators. The average image cross-correlation coefficient obtained using the proposed algorithm was improved by 6-8%. 3D elastographic simulations conducted to study the performance of a 3D elastographic imaging framework predict achievable axial and lateral resolutions of approximately five and ten wavelengths, respectively. A close correspondence between inclusions reconstructed from experimental elastograms and the known physical shape of actual 3D inclusions demonstrates the potential application of 3D elastography for identifying and classifying the detected lesions (invisible in sonograms) on the basis of their shape.

High frequency ultrasound imaging detects cardiac dyssynchrony in noninfarcted regions of the murine left ventricle late after reperfused myocardial infarction.

Cardiac dyssynchrony in the left ventricles of murine hearts late (> or =28 d) after reperfused myocardial infarction (post-MI) was assessed using high frequency 30 MHz B-mode ultrasound imaging. Nine post-MI and six normal C57Bl/6 mice were studied in both short- and long-axis views. Regional time to peak displacement (T(peak_d)) and time to peak strain (T(peak_s)) were calculated in 36 sectors along the myocardial circumference; then their standard deviations (SD_T(peak_d) and SD_T(peak_s)) were computed among noninfarcted myocardial regions for each mouse and were compared between the normal and post-MI mouse groups with Students t-test. The comparison revealed that SD_T(peak_d) and SD_T(peak_s) were significantly larger in the post-MI hearts than in the normal hearts. The displacement uniformity ratio was determined to be 0.97 +/- 0.01 and 0.85 +/- 0.07 for radial and circumferential displacements in the normal hearts, respectively; and 0.59 +/- 0.17 and 0.64 +/- 0.24 in the post-MI hearts. In conclusion, this high resolution ultrasound image tracking method provides for the detection of cardiac dyssynchrony in the noninfarcted regions in the murine left ventricles late after MI by identifying the temporal and spatial disparity of regional myocardial contraction.

Guiding automated left ventricular chamber segmentation in cardiac imaging using the concept of conserved myocardial volume

The active surface technique using gradient vector flow allows semi-automated segmentation of ventricular borders. The accuracy of the algorithm depends on the optimal selection of several key parameters. We investigated the use of conservation of myocardial volume for quantitative assessment of each of these parameters using synthetic and in vivo data. We predicted that for a given set of model parameters, strong conservation of volume would correlate with accurate segmentation. The metric was most useful when applied to the gradient vector field weighting and temporal step-size parameters, but less effective in guiding an optimal choice of the active surface tension and rigidity parameters.

Free-hand ultrasound scanning approaches for volume quantification of the mouse heart left ventricle

Two approaches for free-hand motion tracking that enable volumetric quantification of the murine heart were investigated. One approach used an instrumented, multijointed articulated arm attached to a 14 MHz ultrasound transducer array. A second approach used an E-beam transducer - a modified linear transducer array containing a main imaging array adjacent to three perpendicular tracking arrays. Motion between successive B-mode image frames was computed by tracking image speckle in each tracking array. Both tracking systems produced accurate results in a phantom validation study (4.50% error and 3.75% error for estimates derived using the articulated arm and E-beam tracking techniques, respectively). The tracking approaches also were tested in vivo on three mice. Results were compared to values obtained by mounting each mouse on a micromanipulator, adjusting its position by 0-5-mm increments, and acquiring B-mode images using a high-resolution ultrasound scanner. Left ventricular end diastolic volume (LVEDV) estimates differed from values obtained using the high-resolution scanner by a mean error of 18.2% and 2.60% for eight scans conducted on each of two mice using the articulated arm, and a mean error of 13.6%, 6.53%, and 12.58% for eight scans conducted on each of three mice using the E-beam

3D cardiac motion estimation using RF signal decorrelation

Mouse cardiac ultrasound imaging is generally acquired as sets of 2D B-mode video or RF data. The high signal bandwidth and frame rate (>100 Hz) required for real-time 3D mouse heart scanning presents a major challenge making it unlikely that direct capture of finely sampled real-time 3D data will be achieved in the near future. Collecting and registering image sets from intersecting orthogonal 2D scan planes enables the estimation of 3D motion, but only at points along lines of intersection between acquired image frames. We propose the use of RF signal decorrelation to estimate elevational motion from 2D data at points other than the lines of intersection. RF data was collected and processed using a 30 MHz VisualSonics Vevo 770 scanner. A mouse heart was scanned along short and long axis imaging planes, producing RF data throughout the entire cardiac cycle. RF decorrelations were computed along all A-lines. Orthogonal displacement measurements were used to compute functions which map decorrelation to displacement at lines of intersection between planes. Combining in-plane speckle tracked estimates and out of plane decorrelation based displacement estimates yielded a full 3D displacement data set for the entire cardiac cycle. Displacements estimated using RF decorrelation were highly correlated with independent (orthogonal) speckle-tracked estimates. Out-of-plane decorrelation provides a method for computing 3D displacements from 2D scan-planes of in vivo mouse heart RF data. 3D displacement vectors may be used to compute 3D strain that may be useful when analyzing ischemic, normal and ldquoborder zonerdquo regions post myocardial infarct.

High resolution 2D quantification of myocardial motion abnormalities in mice using high resolution ultrasound with MRI validation

High resolution ultrasound image sequences of left-ventricular motion abnormalities in mice were analyzed on a regional basis using an optimized 2D pixel block tracking technique. The mouse heart was imaged at 35 MHz with approximately 40 mum axial resolution and 100 mum lateral resolution. Myocardial regional motion was tracked using high frame rate mouse heart image sequences using 2D minimum sum of absolute differences (MSAD) block matching. Eight pixel parallel processing was achieved using single instruction multiple data (SIMD) instructions, resulting in a processing speed of 0.11 seconds per frame on a 2.9 GHz Pentium4 PC. Significant differences were observed in the regional displacement, strain and velocity between healthy and infarcted myocardial tissue. Myocardial strains were compared to those obtained via magnetic resonance imaging (MRI). The correlation between the assessment using ultrasound and MRI was R = 91% for radial strain and R = 84% for circumferential strain. Thus, high resolution ultrasound followed by 2D image motion post-processing offers the potential to derive quantitative measures of mouse heart function with the low cost, fine temporal resolution and ease-of-mobility of ultrasound instrumentation

Left Ventricle Segmentation Using Model Fitting and Active Surfaces

A method to perform 4D (3D over time) seg mentation of the left ventricle of a mouse heart using a set of B mode cine slices acquired in vivo from a series of short axis scans is described. We incorporate previ ously suggested methods such as temporal propagation, the gradient vector flow active surface, superquadric models, etc. into our proposed 4D segmentation of the left ventricle. The contributions of this paper are incor poration of a novel despeckling method and the use of locally fitted superellipsoid models to provide a better initialization for the active surface segmentation algorithm. Average distances of the improved surface segmentation to a manually segmented surface through out the entire cardiac cycle and cross-sectional contours are provided to demonstrate the improvements pro duced by the proposed 4D segmentation.

A four-dimensional model-based method for assessing cardiac contractile dyssynchrony in mice

Four-dimensional (4D), or equivalently, 3D + time, analysis is useful for comprehensive assessment cardiac function, especially in the asymmetric left ventricle (LV) after myocardial infarction (MI). This paper presents a 4D-model-based method for ultrasound assessment of cardiac contractile function in mice. Echocardiographic image sequences were acquired at high frequency (30 MHz) from the hearts of C57Bl/6 mice. Image sequences were acquired at contiguous slice locations encompassing the entire 3D LV. In order to reconstruct continuous, dynamic 3D LVs from the images using a 4D mathematical cardiac model, endocardial and epicardial contours were segmented for all image slice locations through one cardiac cycle. In the 4D model, shape and continuity constraints were applied in order to normalize irregularities caused by noise or non-uniform distribution of image data. Root mean square error (RMSE) was calculated between the model-fitted 4D LV and the actual LV surface measured from image data. RMSE was 0.23 mm (~4.5% of epicardial diameter) for the epicardial surface, and 0.20 mm (6.4% of endocardial diameter) for the endocardial surface. 3D regional wall thickening was calculated from the 4D LV surface, and LV dyssynchrony was assessed by analyzing the time to peak strain (Tpeak). This 3D analysis of contractile function in post-MI mouse hearts revealed >80% reduction of peak radial displacement, a 10-15 ms delay in Tpeak in the infarct zone, and a SD_Tpeak of 6-10 ms over the entire 3D LV. In summary, the 4D-model-based method was successfully used for analyzing cardiac dyssynchrony in the 3D murine LV, and it proved advantageous over conventional 2D methods because it was more comprehensive and noise-robust.

The application of the principle of conserved myocardium volume in guiding automated chamber estimation in mouse cardiac imaging

Active contours have been used in a wide variety of image processing applications due to their ability to effectively distinguish image boundaries with limited user input. In this paper, we consider 3D gradient vector field (GVF) active surfaces and their application in the determination of the volume of the mouse heart left ventricle. The accuracy and efficacy of a 3D active surface is strongly dependent upon the selection of several parameters, corresponding to the tension and rigidity of the active surface and the weight of the GVF. However, selection of these parameters is often subjective and iterative. We observe that the volume of the cardiac muscle is, to a good approximation, conserved through the cardiac cycle. Therefore, we propose using the degree of conservation of heart muscle volume as a metric for assessing optimality of a particular set of active surface parameters. A synthetic dataset consisting of nested ellipsoids of known volume was constructed. The outer ellipsoid contracted over time to imitate a heart cycle, and the inner ellipsoid compensated to maintain constant volume. The segmentation algorithm was also investigated in vivo using B-mode data sets obtained by scanning the hearts of three separate mice. Active surfaces were initialized using a broad range of values for each of the parameters under consideration. Conservation of volume was a useful predictor of the efficacy of the model for the range of values tested for the GVF weighting parameter, though it was less effective at predicting the efficacy of the active surface tension and rigidity parameters.