Simulating coronary arteries in x-ray angiograms

Evaluation of internal noise methods for Hotelling observer models

The inclusion of internal noise in model observers is a common method to allow for quantitative comparisons between human and model observer performance in visual detection tasks. In this article, we studied two different strategies for inserting internal noise into Hotelling model observers. In the first strategy, internal noise was added to the output of individual channels: (a) Independent nonuniform channel noise, (b) independent uniform channel noise. In the second strategy, internal noise was added to the decision variable arising from the combination of channel responses. The standard deviation of the zero mean internal noise was either constant or proportional to: (a) the decision variable's standard deviation due to the external noise, (b) the decision variable's variance caused by the external noise, (c) the decision variable magnitude on a trial to trial basis. We tested three model observers: square window Hotelling observer (HO), channelized Hotelling observer (CHO), and Laguerre-Gauss Hotelling observer (LGHO) using a four alternative forced choice (4AFC) signal known exactly but variable task with a simulated signal embedded in real x-ray coronary angiogram backgrounds. The results showed that the internal noise method that led to the best prediction of human performance differed across the studied model observers. The CHO model best predicted human observer performance with the channel internal noise. The HO and LGHO best predicted human observer performance with the decision variable internal noise. The present results might guide researchers with the choice of methods to include internal noise into Hotelling model observers when evaluating and optimizing medical image quality.

Dual-energy mammography: simulation studies

This paper presents a mammography simulator and demonstrates its applicability in feasibility studies in dual-energy (DE) subtraction mammography. This mammography simulator is an evolution of a previously presented x-ray imaging simulation system, which has been extended with new functionalities that are specific for DE simulations. The new features include incident exposure and dose calculations, the implementation of a DE subtraction algorithm as well as amendments to the detector and source modelling. The system was then verified by simulating experiments and comparing their results against published data. The simulator was used to carry out a feasibility study of the applicability of DE techniques in mammography, and more precisely to examine whether this modality could result in better visualization and detection of microcalcifications. Investigations were carried out using a 3D breast software phantom of average thickness, monoenergetic and polyenergetic beam spectra and various detector configurations. Dual-shot techniques were simulated. Results showed the advantage of using monoenergetic in comparison with polyenergetic beams. Optimization studies with monochromatic sources were carried out to obtain the optimal low and high incident energies, based on the assessment of the figure of merit of the simulated microcalcifications in the subtracted images. The results of the simulation study with the optimal energies demonstrated that the use of the DE technique can improve visualization and increase detectability, allowing identification of microcalcifications of sizes as small as 200 microm. The quantitative results are also verified by means of a visual inspection of the synthetic images.

Task-based model/human observer evaluation of SPIHT wavelet compression with human visual system-based quantization

RATIONALE AND OBJECTIVE

The set partitioning in hierarchical trees (SPIHT) wavelet image compression algorithm with the human visual system (HVS) quantization matrix was investigated using x-ray coronary angiograms. We tested whether the HVS quantization matrix for the SPIHT wavelet compression improved computer model/human observer performance in a detection task with variable signals compared to performance with the default quantization matrix. We also tested the hypothesis of whether evaluating the rank order of the two quantization matrices (HVS versus default) based on performance of computer model observers in a signal known exactly but variable task (SKEV) generalized to model/human performance in the more clinically realistic signal known statistically task (SKS).

MATERIALS AND METHODS

Nine hundred test images were created using real x-ray coronary angiograms as backgrounds and simulated arteries with filling defects (signals). The task for the model and human observer was to detect which one of the four computer simulated arterial segments contained the signal, four alternative-forced-choice (4 AFC). We obtained performance for four model observers (nonprewhitening matched filter with an eye filter, Hotelling, Channelized Hotelling, and Laguerre Gauss Hotelling model observers) for both the SKEV and SKS tasks with images compressed with and without the HVS quantization matrix. A psychophysical study measured performance from three human observers for the same conditions and tasks as the model observers.

RESULTS

Performance for all four model observers improved with the use of the HVS quantization scheme. Improvements ranged from 5% (at compression ratio 7:1) to 50% (at compression ratio 30:1) for both the SKEV and SKS tasks. Human observer performance improvement averaged across observers ranged from 6% (at compression ratio 7:1) to 35% (at compression ratio 30:1) for the SKEV task and from 2% (at compression ratio 7:1) to 38% (at compression ratio 30:1) for the SKS task. Addition of internal noise to the model observers allowed for good prediction of human performance.

CONCLUSIONS

Use of the HVS quantization scheme in the SPIHT wavelet compression led to improved model and human observer performance in clinically relevant detection tasks in x-ray coronary angiograms. Model observer performance can be reliably used to predict the human observer performance for the studied tasks as a function of SPIHT wavelet image compression. Our results further confirmed that model observer performance in the computationally more tractable SKEV task can be potentially used as a figure of merit for the more clinically realistic SKS task with real anatomic backgrounds.

Evaluation of JPEG 2000 encoder options: human and model observer detection of variable signals in X-ray coronary angiograms

Previous studies have evaluated the effect of the new still image compression standard JPEG 2000 using nontask based image quality metrics, i.e., peak-signal-to-noise-ratio (PSNR) for nonmedical images. In this paper, the effect of JPEG 2000 encoder options was investigated using the performance of human and model observers (nonprewhitening matched filter with an eye filter, square-window Hotelling, Laguerre-Gauss Hotelling and channelized Hotelling model observer) for clinically relevant visual tasks. Two tasks were investigated: the signal known exactly but variable task (SKEV) and the signal known statistically task (SKS). Test images consisted of real X-ray coronary angiograms with simulated filling defects (signals) inserted in one of the four simulated arteries. The signals varied in size and shape. Experimental results indicated that the dependence of task performance on the JPEG 2000 encoder options was similar for all model and human observers. Model observer performance in the more tractable and computationally economic SKEV task can be used to reliably estimate performance in the complex but clinically more realistic SKS task. JPEG 2000 encoder settings different from the default ones resulted in greatly improved model and human observer performance in the studied clinically relevant visual tasks using real angiography backgrounds.

Automated optimization of JPEG 2000 encoder options based on model observer performance for detecting variable signals in X-ray coronary angiograms

Image compression is indispensable in medical applications where inherently large volumes of digitized images are presented. JPEG 2000 has recently been proposed as a new image compression standard. The present recommendations on the choice of JPEG 2000 encoder options were based on nontask-based metrics of image quality applied to nonmedical images. We used the performance of a model observer [non-prewhitening matched filter with an eye filter (NPWE)] in a visual detection task of varying signals [signal known exactly but variable (SKEV)] in X-ray coronary angiograms to optimize JPEG 2000 encoder options through a genetic algorithm procedure. We also obtained the performance of other model observers (Hotelling, Laguerre-Gauss Hotelling, channelized-Hotelling) and human observers to evaluate the validity of the NPWE optimized JPEG 2000 encoder settings. Compared to the default JPEG 2000 encoder settings, the NPWE-optimized encoder settings improved the detection performance of humans and the other three model observers for an SKEV task. In addition, the performance also was improved for a more clinically realistic task where the signal varied from image to image but was not known a priori to observers [signal known statistically (SKS)]. The highest performance improvement for humans was at a high compression ratio (e.g., 30:1) which resulted in approximately a 75% improvement for both the SKEV and SKS tasks.

Vessel size measurements in angiograms: manual measurements

Vessel size measurement is perhaps the most often performed quantitative analysis in diagnostic and interventional angiography. Although automated vessel sizing techniques are generally considered to have good accuracy and precision, we have observed that clinicians rarely use these techniques in standard clinical practice, choosing to indicate the edges of vessels and catheters to determine sizes and calibrate magnifications, i.e., manual measurements. Thus, we undertook an investigation of the accuracy and precision of vessel sizes calculated from manually indicated edges of vessels. Manual measurements were performed by three neuroradiologists and three physicists. Vessel sizes ranged from 0.1-3.0 mm in simulation studies and 0.3-6.4 mm in phantom studies. Simulation resolution functions had full-widths-at-half-maximum (FWHM) ranging from 0.0 to 0.5 mm. Phantom studies were performed with 4.5 in., 6 in., 9 in., and 12 in. image intensifier modes, magnification factor = 1, with and without zooming. The accuracy and reproducibility of the measurements ranged from 0.1 to 0.2 mm, depending on vessel size, resolution, and pixel size, and zoom. These results indicate that manual measurements may have accuracies comparable to automated techniques for vessels with sizes greater than 1 mm, but that automated techniques which take into account the resolution function should be used for vessels with sizes smaller than 1 mm.

Vessel size measurements in angiograms: a comparison of techniques

As interventional procedures become more complicated, the need for accurate quantitative vascular information increases. In response to this need, many commercial vendors provide techniques for measurement of vessel sizes, usually based on derivative techniques. In this study, we investigate the accuracy of several techniques used in the measurement of vessel size. Simulated images of vessels having circular cross sections were generated and convolved with various focal spot distributions taking into account the magnification. These vessel images were then convolved with Gaussian image detector line spread functions (LSFs). Additionally, images of a phantom containing vessels with a range of diameters were acquired for the 4.5", 6", 9", and 12" modes of an image intensifier-TV (II-TV) system. Vessel sizes in the images were determined using a first-derivative technique, a second-derivative technique, a linear combination of these two measured sizes, a thresholding technique, a densitometric technique, and a model-based technique. For the same focal spot size, the shape of the focal spot distribution does not affect measured vessel sizes except at large magnifications. For vessels with diameters larger than the full-width-at-half-maximum (FWHM) of the LSF, accurate vessel sizes (errors approximately 0.1 mm) could be obtained by using an average of sizes determined by the first and second derivatives. For vessels with diameters smaller than the FWHM of the LSF, the densitometric and model-based techniques can provide accurate vessel sizes when these techniques are properly calibrated.

Evaluation of layer decomposition for multiframe quantitative coronary angiography

Multiframe quantitative coronary angiography is typically performed by averaging measurements of artery diameter over multiple frames. This approach reduces errors attributable to random noise but may not reduce systematic errors caused by background structures, nonlinear system response, and motion blur. We attempt to reduce these sources of error by decomposing the image sequence into moving layers, one of which includes the artery. We embed simulated arteries into clinical angiographic sequences so that the true vessel dimensions are known accurately. The measurement tasks are minimum diameter, geometric percent stenosis, and densitometric percent stenosis. We compare measurements for single and multiple raw images, single images with fixed mask subtraction, single and multiple images with layered background subtraction, and time-averaged layer images. We find that both multiframe averaging and layer decomposition significantly improve geometric and densitometric accuracy compared with single-frame measurements. The best results were obtained by averaging measurements from multiple frames of layered background-subtracted images.

Accuracy assessment of layer decomposition using simulated angiographic image sequences

Layer decomposition is a promising method for obtaining accurate densitometric profiles of diseased coronary artery segments. This method decomposes coronary angiographic image sequences into moving densitometric layers undergoing translation, rotation, and scaling. In order to evaluate the accuracy of this technique, we have developed a technique for embedding realistic simulated moving stenotic arteries in real clinical coronary angiograms. We evaluate the accuracy of layer decomposition in two ways. First, we compute tracking errors as the distance between the true and estimated motion of a reference point in the arterial lesion. We find that noise-weighted phase correlation and layered background subtraction are superior to cross correlation and fixed mask subtraction, respectively. Second, we compute the correlation coefficient between the true vessel profile and the raw and processed images in the region of the stenosis. We find that layer decomposition significantly improves the correlation coefficient.