An estimated 17.5 million people died from a cardiovascular disease in 2012, representing 31% of all global deaths. Most acute coronary events result from rupture of the protective fibrous cap overlying an atherosclerotic plaque. The task of early identification of plaque types that can potentially rupture is, therefore, of great importance. The state-ofthe-art approach to imaging blood vessels is intravascular optical coherence tomography (IVOCT). However, currently, this is an offline approach where the images are first collected and then manually analyzed a frame at a time to identify regions at risk of thrombosis. This process is extremely laborious, time consuming and prone to human error. We are building a system that, when complete, will provide interactive 3D visualization of a blood vessel as an IVOCT is in progress. The visualization will highlight different plaque types and enable quick identification of regions at risk for thrombosis. In this paper, we describe our approach, focusing on machine learning methods that are a key enabling technology. Our empirical results using real OCT data show that our approach can identify different plaque types efficiently with high accuracy across multiple patients.

With its single cell sensitivity over volumes as large as or larger than a mouse, cryo-imaging enables imaging of stem cell biodistribution, homing, engraftment, and molecular mechanisms. We developed and evaluated a highly automated software tool to detect fluorescently labeled stem cells within very large ( ∼ 200 GB) cryo-imaging datasets. Cell detection steps are: preprocess, remove immaterial regions, spatially filter to create features, identify candidate pixels, classify pixels using bagging decision trees, segment cell patches, and perform 3D labeling. There are options for analysis and visualization. To train the classifier, we created synthetic images by placing realistic digital cell models onto cryo-images of control mice devoid of cells. Very good cell detection results were (precision=98.49%, recall=99.97%) for synthetic cryo-images, (precision=97.81%, recall=97.71%) for manually evaluated, actual cryo-images, and false positives in control mice. An α-multiplier applied to features allows one to correct for experimental variations in cell brightness due to labeling. On dim cells (37% of standard brightness), with correction, we improved recall (49.26%→ 99.36%) without a significant drop in precision (99.99%→ 99.75%) . With tail vein injection, multipotent adult progenitor cells in a graft-versus-host-disease model in the first days post injection were predominantly found in lung, liver, spleen, and bone marrow. Distribution was not simply related to blood flow. The lung contained clusters of cells while other tissues contained single cells. Our methods provided stem cell distribution anywhere in mouse with single cell sensitivity. Methods should provide a rational means of evaluating dosing, delivery methods, cell enhancements, and mechanisms for therapeutic cells.

Worldwide, many hundreds of thousands of stents are implanted each year to revascularize occlusions in coronary arteries. Intravascular optical coherence tomography is an important emerging imaging technique, which has the resolution and contrast necessary to quantitatively analyze stent deployment and tissue coverage following stent implantation. Automation is needed, as current, it takes up to 16 h to manually analyze hundreds of images and thousands of stent struts from a single pullback. For automated strut detection, we used image formation physics and machine learning via a Bayesian network, and 3-D knowledge of stent structure via graph search. Graph search was done on en face projections using minimum spanning tree algorithms. Depths of all struts in a pullback were simultaneously determined using graph cut. To assess the method, we employed the largest validation data set used so far, involving more than 8000 clinical images from 103 pullbacks from 72 patients. Automated strut detection achieved a 0.91±0.04 recall, and 0.84±0.08 precision. Performance was robust in images of varying quality. This method can improve the workflow for analysis of stent clinical trial data, and can potentially be used in the clinic to facilitate real-time stent analysis and visualization, aiding stent implantation.

Recent innovations in 3D motion capture technology such as electromagnetic articulography (EMA) are providing unprecedented access to the intricate movements of the articulators during speech production. Although these technological advances afford exciting opportunities for advancing the assessment and treatment of speech, they have presented new challenges associated with data collection, processing, and analysis. To address these challenges, we have standardized our EMA data collection protocols and developed a Matlab-based software tool, SMASH, for processing, visualizing, and analyzing speech movement data. The goal of the software is to advance research on speech production by improving the efficiency and reliability of speech movement analyses.

Decomposition of water and fat in magnetic resonance imaging (MRI) is important for biomedical research and clinical applications. In this paper, we propose a two-phased approach for the three-point water-fat decomposition problem. Our contribution consists of two components: 1) a background-masked Markov random field (MRF) energy model to formulate the local smoothness of field inhomogeneity; 2) a new iterated conditional modes (ICM) algorithm accounting for high-performance optimization of the MRF energy model. The MRF energy model is integrated with background masking to prevent error propagation of background estimates as well as improve efficiency. The central component of our new ICM algorithm is the stability tracking (ST) mechanism intended to dynamically track iterative stability on pixels so that computation per iteration is performed only on instable pixels. The ST mechanism significantly improves the efficiency of ICM. We also develop a median-based initialization algorithm to provide good initial guesses for ICM iterations, and an adaptive gradient-based scheme for parametric configuration of the MRF model. We evaluate the robust of our approach with high-resolution mouse datasets acquired from 7T MRI.

We are developing methods to characterize atherosclerotic disease in human carotid arteries using multiple MR images having different contrast mechanisms (T1W, T2W, PDW). To enable the use of voxel gray values for interpretation of disease, we created a new method, local entropy minimization with a bicubic spline model (LEMS), to correct the severe (approximately 80%) intensity inhomogeneity that arises from the surface coil array. This entropy-based method does not require classification and robustly addresses some problems that are more severe than those found in brain imaging, including noise, steep bias field, sensitivity of artery wall voxels to edge artifacts, and signal voids near the artery wall. Validation studies were performed on a synthetic digital phantom with realistic intensity inhomogeneity, a physical phantom roughly mimicking the neck, and patient carotid artery images. We compared LEMS to a modified fuzzy c-means segmentation based method (mAFCM), and a linear filtering method (LINF). Following LEMS correction, skeletal muscles in patient images were relatively isointense across the field of view. In the physical phantom, LEMS reduced the variation in the image to 1.9% and across the vessel wall region to 2.5%, a value which should be sufficient to distinguish plaque tissue types, based on literature measurements. In conclusion, we believe that the correction method shows promise for aiding human and computerized tissue classification from MR signal intensities.

This paper presents a use of geometric-based method integrated with classifier for lumen wall estimation using MR plaque volumes. The following are the new things the readers will observe when it comes to plaque imaging. (a) Application of three different sets of : classifiers (fuzzy, Markovian and graph-based) for lumen region classification in plaque MR volumes. These classifiers are used in multi-resolution framework. (b) Usage of rule-based region merging applied to the sub-classes of lumen region. (c) Rotational effect on region of interest in arterial bifurcation zones for accurate lumen region identification and boundary estimation. We have used our diagnostic system with three different classifying methods on actual patient data. We measure performance of the system by computing the mean distance error with respect to boundaries traced manually by human experts. Overall, the system consists of 22,500 boundary points. The in-plane pixel resolution is 0.25 millimeters. Using Markovian classifier method, the average error was 0.61 pixels; using fuzzy classifier method, the average error was 0.62 pixels; using graph-based classifier method, the average error was 0.74 pixels. All these methods lead to error less than 0.185 mm. We also validated our system by simulating the lumen images with additive Gaussian perturbations. This system works on a Linux platform and is written in C++.

In this study, we registered live-time interventional magnetic resonance imaging (iMRI) slices with a previously obtained high-resolution MRI volume that in turn can be registered with a variety of functional images, e.g., PET, SPECT, for tumor targeting. We created and evaluated a slice-to-volume (SV) registration algorithm with special features for its potential use in iMRI-guided radio-frequency (RF) thermal ablation of prostate cancer. The algorithm features included a multiresolution approach, two similarity measures, and automatic restarting to avoid local minima. Imaging experiments were performed on volunteers using a conventional 1.5-T MR scanner and a clinical 0.2-T C-arm iMRI system under realistic conditions. Both high-resolution MR volumes and actual iMRI image slices were acquired from the same volunteers. Actual and simulated iMRI images were used to test the dependence of SV registration on image noise, receive coil inhomogeneity, and RF needle artifacts. To quantitatively assess registration, we calculated the mean voxel displacement over a volume of interest between SV registration and volume-to-volume registration, which was previously shown to be quite accurate. More than 800 registration experiments were performed. For transverse image slices covering the prostate, the SV registration algorithm was 100% successful with an error of <2 mm, and the average and standard deviation was only 0.4 mm +/- 0.2 mm. Visualizations such as combined sector display and contour overlay showed excellent registration of the prostate and other organs throughout the pelvis. Error was greater when an image slice was obtained at other orientations and positions, mostly because of inconsistent image content such as that from variable rectal and bladder filling. These preliminary experiments indicate that MR SV registration is sufficiently accurate to aid image-guided therapy.

We created a method for three-dimensional (3-D) registration of medical images (e.g., magnetic resonance imaging (MRI) or computed tomography) to images of physical tissue sections or to other medical images and evaluated its accuracy. Our method proved valuable for evaluation of animal model experiments on interventional-MRI guided thermal ablation and on a new localized drug delivery system. The method computes an optimum set of rigid body registration parameters by minimization of the Euclidean distances between automatically chosen correspondence points, along manually selected fiducial needle paths, and optional point landmarks, using the iterative closest point algorithm. For numerically simulated experiments, using two needle paths over a range of needle orientations, mean voxel displacement errors depended mostly on needle localization error when the angle between needles was at least 20 degrees. For parameters typical of our in vivo experiments, the mean voxel displacement error was < 0.35 mm. In addition, we determined that the distance objective function was a useful diagnostic for predicting registration quality. To evaluate the registration quality of physical specimens, we computed the misregistration for a needle not considered during the optimization procedure. We registered an ex vivo sheep brain MR volume with another MR volume and tissue section photographs, using various combinations of needle and point landmarks. Mean registration error was always < or = 0.54 mm for MR-to-MR registrations and < or = 0.52 mm for MR to tissue section registrations. We also applied the method to correlate MR volumes of radio-frequency induced thermal ablation lesions with actual tissue destruction. In this case, in vivo rabbit thigh volumes were registered to photographs of ex vivo tissue sections using two needle paths. Mean registration errors were between 0.7 and 1.36 mm over all rabbits, the largest error less than two MR voxel widths. We conclude that our method provides sufficient spatial correspondence to facilitate comparison of 3-D image data with data from gross pathology tissue sections and histology.

Recent advances in drug delivery techniques have necessitated the development of tools for in vivo monitoring of drug distributions. Gamma emission imaging and magnetic resonance imaging suffer from problems of resolution and sensitivity, respectively. We propose that the combination of X-ray CT imaging and image analysis techniques provides an excellent method for the evaluation of the transport of platinum-containing drugs from a localized, controlled release source. We correlated local carboplatin concentration with CT intensity, producing a linear relationship with a sensitivity of 62.6 microg/mL per Hounsfield unit. As an example application, we evaluated the differences in drug transport properties between normal and ablated rabbit liver from implanted polymer millirods. The use of three-dimensional visualization provided a method of evaluating the placement of the drug delivery device in relation to the surrounding anatomy, and registration and reformatting allowed the accurate comparison of the sequence of temporal CT volumes acquired over a period of 24 h. Taking averages over radial lines extending away from the center of the implanted millirods and integrating over clinically appropriate regions, yielded information about drug release from the millirod and transport in biological tissues. Comparing implants in normal and ablated tissues, we found that ablation prior to millirod implantation greatly decreased the loss of drug from the immediate area, resulting in a higher average dose to the surrounding tissue. This work shows that X-ray CT imaging is a useful technique for the in vivo evaluation of the pharmacokinetics of platinated agents.

Digital temporal and spatial filtering of fluoroscopic image sequences can be used to improve the quality of images acquired at low X-ray exposure. In this study, we characterized a nonlinear edge preserving, spatio-temporal noise reduction filter, the bidirectional multistage (BMS) median filter of Arce (1991). To assess image quality, signal detection and discrimination experiments were performed on stationary targets using a four-alternative forced-choice paradigm. A measure of detectability, d', was obtained for filtered and unfiltered noisy image sequences at different signal amplitudes. Filtering gave statistically significant, average d' improvements of 20% (detection) and 31% (discrimination). A nonprewhitening detection model modified to include the human spatio-temporal visual system contrast-sensitivity underestimated enhancement, predicting an improvement of 6%. Pixel noise standard deviation, a commonly applied image quality measure, greatly overestimated effectiveness giving 67% improvement in d'. We conclude that human testing is required to evaluate the filter effectiveness and that human perception models must be improved to account for the spatio-temporal filtering of image sequences.

Using a one-dimensional convective-dispersive model of contrast agent flow in a blood vessel, we optimized and compared algorithms for combining a temporal sequence of X-ray angiography images, each with incomplete arterial filling, into a single-output image with fully opacified arteries. The four algorithms were: maximum opacity (MO) with a maximum over time at each spatial location; matched filtering (MAT); recursive filtering (REC) with a maximum opacity; and an approximate matched filter (AMF) consisting of a correlation with a kernel that approximates the matched filter kernel followed by a maximum opacity operation. Based on the contrast-to-noise ratio (CNR), MAT is theoretically the best algorithm. However, with spatially varying clinical images, a poorly matched MAT kernel greatly degraded CNR to the point of even inverting artery contrast. The practical AMF method maintained uniform CNR values over the entire field of view and gave >90% of the theoretical limit set by MAT. REC and MO created fully opacified arteries, but provided little CNR enhancement. By holding CNR at a nominal reference value, simulations predicted that AMF could be used with a contrast agent volume reduced by as much as 66%. Alternatively, X-ray exposure rate could be lowered. Although MO and REC are more easily implemented, the contrast enhancement with AMF makes it attractive for processing diagnostic angiography images acquired with a reduced contrast agent dose.

We evaluated semiautomatic, voxel-based registration methods for a new application, the assessment and optimization of interventional magnetic resonance imaging (I-MRI) guided thermal ablation of liver cancer. The abdominal images acquired on a low-field-strength, open I-MRI system contain noise, motion artifacts, and tissue deformation. Dissimilar images can be obtained as a result of different MRI acquisition techniques and/or changes induced by treatments. These features challenge a registration algorithm. We evaluated one manual and four automated methods on clinical images acquired before treatment, immediately following treatment, and during several follow-up studies. Images were T2-weighted, T1-weighted Gd-DTPA enhanced, T1-weighted, and short-inversion-time inversion recovery (STIR). Registration accuracy was estimated from distances between anatomical landmarks. Mutual information gave better results than entropy, correlation, and variance of gray-scale ratio. Preprocessing steps such as masking and an initialization method that used two-dimensional (2-D) registration to obtain initial transformation estimates were crucial. With proper preprocessing, automatic registration was successful with all image pairs having reasonable image quality. A registration accuracy of approximately equal to 3 mm was achieved with both manual and mutual information methods. Despite motion and deformation in the liver, mutual information registration is sufficiently accurate and robust for useful applications in I-MRI thermal ablation therapy.

Stereoscopy can be an effective method for obtaining three-dimensional (3-D) spatial information from two-dimensional (2-D) projection X-ray images, without the need for tomographic reconstruction. This much-needed information is missed in many X-ray diagnostic and interventional procedures, such as the treatment of vascular aneurysms. Fast C-arm X-ray systems can obtain multiple angle sequences of stereoscopic image pairs from a single contrast injection and a single breath hold. To advance this solution, we developed a model of stereo angiography, performed perception experiments and related results to optimal acquisition. The model described horizontal disparity for the C-arm geometry that agreed very well with measurements from a geometric phantom. The perceptual accommodation-convergence conflict and geometry limited the effective stereoscopic field of view (SFOV). For a typical large image intensifier system, it was 28 cm x 31 cm at the center of rotation (COR). In the model, blurring from finite focal-spot size and C-arm motion reduced depth resolution on the digital display. Near the COR, the predicted depth resolution was 3-11 mm for a viewing angle of 7 degrees , which agreed favorably with results from recently published studies. The model also described how acquisition parameters affected spatial warping of curves of equal apparent depth. Pincushioning and the difference between the acquisition and display geometry were found to introduce additional distortions to stereo displays. Preference studies on X-ray angiograms indicated that the ideal viewing angle should be small (1-2 degrees), which agreed with some previously published work. Perceptual studies indicated that stereo angiograms should have high artery contrast and that digital processing to increase contrast improved stereopsis. Digital subtraction angiograms, with different motion errors between the left and right-eye views, gave artifacts that confused stereopsis. The addition of background to subtracted images reduced this effect and provided other features for improved depth perception. Using the modeling results and typical clinical angiography requirements, we recommend acquisition protocols and engineering specifications that are achievable on current high-end systems.

For noisy X-ray fluoroscopy image sequences we quantitatively evaluated image quality after digital temporal filtering to reduce noise. Using an experimental paradigm called a reference/test adaptive forced-choice method we compared detectability of stationary low-contrast disks in filtered and unfiltered, computer-generated image sequences. In the first experiment, a low-pass first-order recursive filter used in X-ray fluoroscopy was found to be much less effective at enhancing detectability than predicted from the reduction of display noise variance, a common measurement of filter effectiveness. Detectability was reasonably predicted by a nonprewhitening human-observer model (NPW-HVS) that included an independently determined human temporal-contrast-sensitivity function. In another experiment, designed to test models over a range of temporal frequencies, we used paired high-pass and low-pass temporal filters that both reduced noise variance by 25%. The high-pass filter was artificially applied to the noise only and greatly improved detectability, while the low-pass filter had little effect. The human-observer model quantitatively described the measurements, but classical prewhitening and nonprewhitening signal detectors did not. As compared to the nonprewhitening, spatio-temporal matched filter, human-observer efficiency was low and variable at 2.1%, 2.9%, and 0.06% for 60 frames of unfiltered low-pass and high-pass noise, respectively. As compared to this detector, humans were not very effective at combining information across frames. On the other hand, signal to noise ratios (SNR's) from the human-observer model were comparable to human performance, and efficiencies were reasonably constant at 40%, 52%, and 32%, respectively. We conclude that it is imperative to include human-observer models and experiments in the analysis of noise-reduction filtering of noisy image sequences, such as X-ray fluoroscopy.

One potential way to reduce patient and staff X-ray fluoroscopy dose is to reduce the quantum exposure to the detector and compensate the additional noise with digital filtering. A new filtering method, spatio-temporal filtering with object detection, is described that reduces noise while minimizing motion and spatial blur. As compared to some conventional motion-detection filtering schemes, this object-detection method incorporates additional a priori knowledge of image content; i.e. much of the motion occurs in isolated long thin objects (catheters, guide wires, etc.). We create object-likelihood images and use these to control spatial and recursive temporal filtering such as to reduce blurring the objects of interest. We use automatically computed receiver operating characteristic (ROC) curves to optimize the object-likelihood enhancement method and determine that oriented matched filter kernels with 4 orientations are appropriate. The matched filter kernels are simple projected cylinders. We demonstrate the method on several representative X-ray fluoroscopy sequences to which noise is added to simulate very low dose acquisitions. With processing, we find that noise variance is significantly reduced with slightly less noise reduction near moving objects. We estimate an effective exposure reduction greater than 80%

We report real time fractal coding of 176 by 144 luminance images on a 33 MHz i-486 computer. With a zero searching first order (bilinear) Bath fractal transform, square blocks of 64 pixels are coded in 250 /spl mu/sec and decoded in 200 /spl mu/sec. This permits coding of 40% of each frame in a 25 Hz video sequence, or full coding at 10 Hz. By transmitting only those blocks of greatest error, we can obtain any degree of data compression. Useful video sequences of talking heads have been achieved with rates as low as 40 kbits/sec. We describe the techniques used, give illustrations of the degradation, and show the evolution of the errors in the decoded sequence as various events occur. The overall compression at 40 kbits/sec is 99:1, with further error reductions possible.>