Open Access
Add to BibSonomyAbstract
We evaluate the effect on human visual detection of a display method that stabilizes the motion of the feature of interest in sequential JPEG compressed x-ray coronary angiogram images. At all levels of image compression the feature stabilized display significantly improved performance with respect to the standard display where the artery is moving. In addition, for both the moving artery and stabilized display, human performance with images compressed at 15:1 was not significantly different from performance with the uncompressed images.
©1999 Optical Society of America
1. Introduction
Cardiac catherization laboratories are quickly moving towards digital angiographic technology that allows immediate review, quantitative analysis and image processing of the coronary angiograms [1]. Although there has been great progress in computer quantitative analysis of x-ray coronary angiograms, visual assessment of the disease by the physician still remains the most important diagnostic procedure. In this context, the digital format readily allows for any image processing and/or manipulation of the display methods that could improve the physician's visual detection, classification and/or estimation of a lesion.
For example, psychophysical studies have investigated visual detection of signals embedded in spatially and temporally varying white noise. In this case, detection performance for an observer that temporally integrates all the images in the sequence prior to detection will increase as the number of frames in the sequence increases [2]. However, human observers are not able to unlimitedly and optimally integrate through time all images in a sequence [2–6]. Instead, humans have a limited temporal window of integration allowing them to integrate up to 700-1000 ms. of data [2,3] and with decreasing integration efficiency as the number of frames increases [2–4]. Therefore, when noisy image sequences are displayed, increasing the display frame rate increases the number of frames displayed within the human integrating window, allowing for more noisy frames to be more efficiently integrated and therefore improve human detection performance [2].
A potential problem when applying this concept to real noisy dynamic medical images such as x-ray coronary angiograms is that the feature of interest (whether it is the artery itself or a morphological feature within an arterial segment) is moving. As a result, as the display frame rate is increased the motion becomes more rapid and abrupt, making visual tracking by the human observer difficult. The difficulty in visual tracking can have detrimental effects on human performance [7]. Eigler et al. [7] proposed an optimized display for coronary angiograms where each image of the sequence is digitally shifted so that the feature of interest within an artery remains fixed at the center of the screen while the background moves. This stabilized display (S.D.) permits fast display of the image sequences while allowing the observer to scrutinize the feature of interest without having to eye-track rapid and erratic arterial motion.
Figure 1. Left standard moving artery display (M.A). Right the stabilized display (S.D.) [Media 1] [Media 2] [Media 3] [Media 4]
Figure 1 (left) shows a sub-region of a 30-frame x-ray coronary angiogram displayed in the standard method where the artery moves back and forth (moving artery, M.A.). Figure 1 (right) shows the same sub-region but with the arterial segment stabilized. Previous results using computer simulated arteries have shown that the S.D. improves feature detectability over the standard display where the artery is moving (M.A.). The goal of the present work is to determine whether the performance improvement with the stabilized display generalizes to JPEG compressed images. In addition, previous work assessing the usefulness of the stabilized display used computer simulated arterial motion [7]. We seek to generalize previous findings to real arterial motion that were obtained by measuring the 2-D image motion of real arteries in x-ray coronary angiograms. Finally, previous studies have shown that 15:1 image compression does not significantly degrade visual detection performance in dynamically displayed angiograms [8, 9]. It not known whether this result holds for a feature-stabilized display.
2. Methods
2.1 Acquisition of x-ray coronary angiograms
The clinical digital coronary angiograms were acquired at 30 frames/s with a 7-in. image intensifier filed size (Advantx/DXC, General Electric Medical Systems). The parameters used in the image acquisition were a standard automatic exposure control at 0.30 μGy per frame and extended-dynamic-range-enabled video circuitry. The images were digitized with a linear analog amplification and lookup table to achieve a 512 × 512 pixel matrix with a resolution of 0.3 mm/pixel and 256 gray levels.
2.2 Generation of simulated arteries and lesions for an individual frame
The algorithm to create the simulated arterial segments and lesions is intended to mimic the image generation process of x-ray coronary angiograms. The present algorithm is an extension of the one previously used [7] and now includes scattering and veiling glare. Details about the algorithm to generate the computer simulated arteries and lesion (filling defect) are discussed in the Appendix.
For our test-images the projected simulated arteries are 3-D right circular cylinders with a diameter of 12 pixels (3.6 mm), a sinusoidally modulated narrowing in diameter toward the center (minimum diameter of 8 pixels), and a length of 50 pixels (15.0 mm). Four simulated arteries were generated for each test-image and projected 32 pixels apart (Figure 2). The attenuation coefficient, μ, was set to 0.16/mm to produce simulated arteries with the same projected intensity as real angiograms of coronary arteries of the same diameter.
Figure 2. Test image-sequences consisting of four simulated arterial segments (projected 3-D right cylinders with narrowing towards the center). The subtle lesion (a bright spot) is located at the vertical and horizontal center of the bottom right simulated arterial segment. From left to right: Uncompressed moving artery display, uncompressed stabilized display, 30:1 JPEG compressed moving artery and 30:1 JPEG compressed stabilized display. [Media 5] [Media 6] [Media 7] [Media 8]
The target consisted of a simulated filling defect with a hemi-ellipsoidal shape located at the vertical and horizontal center of one of the four simulated arteries with a diameter of 6 pixels (Figure 2). We simulated imaging system blur caused by the physical extent of the x-ray focal spot and image intensifier unsharpness by convolving the projected cylinders with a isotropic Gaussian point spread function with standard deviation of 1 pixel (0.3 mm).
2.3 Generation of arterial motion and feature stabilization
Unlike the test-images used in previous studies [7], the motion for the simulated arterial segments in the present study were obtained from real arteries. The positions of the arteries through a 32-frame sequence were manually tracked. A human operator selected a feature within a given artery in the first frame of the image sequence by locating the cursor on top of the feature and pressing the left button of the mouse. The computer then recorded the x and y pixel coordinates of the feature of interest. The operator then scrutinized each subsequent frame localizing the position of the feature of interest and selecting it with the cursor. At the end of this process, a file was generated that contained the x and y pixel coordinates for the feature of interest within the artery for the 32 frames. We recorded artery motion coordinates for over 40 different angiographic sequences. These x and y coordinates were then used as the positions for the simulated arteries through the 32 frame sequences of simulated arteries. The feature-stabilized sequences were obtained by using the known positions of the simulated arteries and digitally shifting the images by these same displacements.
2.4 JPEG compression of images
Images were compressed and decompressed with the fifth public release of the Independent JPEG Group’s free JPEG software. Since a given quality level results in varying compression ratios for different images, a program was written that would iteratively modify the quality level of the compression until the target compression ratio was achieved (with a tolerance of up to 5 % error). In this way the set of 100 image sequences were compressed to achieve final compression ratios (averaged compression across all image sequences ± standard deviation) of 6.94 ± 0.17, 14.89 ± 0.156 and 29.79 ± 0.295.
2.5 Psychophysical studies
The observer’s task was to detect the filling defect at the vertical and horizontal center of one of four simulated arteries (4 alternative forced choice). In Figure 2, the target is located in the upper left artery. In each trial an image sequence was randomly sampled from the 100 image-set database. There were a total of 8 different experimental conditions. There were 4 compression conditions (uncompressed, 7:1, 15:1 and 30:1 image compression) × 2 display conditions (M.A. vs. S.D.). Figure 2 shows an image sequence shown in the uncompressed moving artery condition (left image sequence) and an image sequence for the uncompressed stabilized condition (second image from left). Figure 2 shows the two display conditions for the 30:1 JPEG compressed condition (from left: third and fourth images). Four observers participated in the experiment. Three of the observers were naïve observers with extended training visually detecting simulated lesions in medical images. The fourth observer (DV) was a radiologist with previous experience in reading x-ray coronary angiograms. Observers first trained in 1 session of 100 trials for each experimental condition and then participated in 5 sessions of 100 trials per condition. Conditions were randomized within a day. Images were displayed on an Image Systems M17LMAX monochrome monitor (Image Systems, Minnetonka, MN). The mean luminance was 16.0 cd/m2. The luminance vs. gray level relationship was the default non-linear curve that would be used by the physicians in a clinic with this monitor. Observers viewed the images binocularly from a distance of 50 cm and had unlimited time to reach a decision. When a decision was reached they pressed the number 1, 2, 3 or 4 in the keyboard to indicate their choice for that trial. The display frame-rates were 16 frames/second for the conventional moving artery display and 32 frames/second for the stabilized display [7]. The choice of a 16 frames/second for the moving artery was based on previous results showing that this frame rate results in better human performance than the 32 frames/second display frame rate. Since our goal is to compare the best-possible conventional display method (moving artery) vs. the stabilized display we chose to use 16 frames/second rather than match the 32 frames/second frame rate used for the stabilized display.
2.6 Data Analysis:
Accuracy for a given observer in a given experimental condition was quantified by computing the percent of trials (Pc) that the observer correctly detected the target. Pc was then transformed to an index of detectability (da) for a 4 alternative forced choice given by [12]:
where , g(z) = 1/√2π exp(-1/2 x 2), G(z) is the cumulative Gaussian, M is the number of alternatives. The index of detectability has been shown to be approximately constant with the number of alternatives in the task and linear with signal contrast for human visual detection in white noise [13] and x-ray coronary angiograms [14, 15].
Figure 3. Human performance (da) for 4 observers detecting a low-contrast feature as a function of JPEG compression ratio for the standard moving artery display (M.A.) and the stabilized display (S.D.). For each graph the highest and lowest performance points are labeled with the corresponding Pc performance level.
3. Results
Figure 3 shows performance as measured by da for each observer for the M.A. and S.D. as a function of increasing JPEG image compression. Performance for all observers is larger for the S.D. display at all levels of compression. There is a reduction in performance with increasing compression for all observers and both display types. A within subject factorial ANOVA resulted in a significant (F(1,3) = 95.31, p < 0.005) effect of display type (S.D. vs. M.A.) and of image compression (F(3,9) = 97.78, p < 0.005) on human performance.
The display type x compression interaction was not significant. Paired comparisons with Bonferroni corrections for multiple comparisons resulted in a significant difference between human performance in the uncompressed condition and 30:1 JPEG compression for both display types. Comparisons between the uncompressed condition and the 7:1 and 15:1 did not reach significance (p > 0.01).
4. Discussion
4.1 Effect of stabilized display
Our results show that the improvement in performance with the stabilized display holds for compressed images. Our results generalize previous findings that used simulated arterial motion to real arterial motion. The % improvement (averaged across observers) for the uncompressed condition found by Eigler et al. [7] for the simulated arterial motion was approximately 50 % compared to about 30 % for the present study. We speculate that the difference might have to do with the different nature of the motions used in the studies (simulated vs. real). Our findings suggest that for the present task, stabilized 15:1 JPEG compressed angiograms result in higher performance than for uncompressed conventionally displayed angiograms.
An assumption in our experimental method is that the locations of the feature of interest can be tracked with no errors, so that the feature is perfectly stabilized. If random errors in the tracking (either by manual tracking by a human operator or a computer algorithm) are present, then the feature of interest will not be perfectly stabilized and instead will have random positional jitter through the sequence. We have previously studied the effect of random Gaussian jitter in the position of the artery, and found that human performance does not degrade for a Gaussian jitter with a standard deviation which is small compared to the feature size but did degrade for larger jitter magnitudes. [16]. Future studies should assess how much position jitter is introduced in the stabilized display by computer tracking algorithms that are currently being developed [17].
4.2 Effect of image compression
Previous studies have assessed the effect of image compression on visual detection of morphological features in x-ray coronary angiograms. However, in these studies the truth about lesion presence was not known and was established by a consensus committee. Recent studies using simulated signals in x-ray coronary angiograms have shown that the “gold standard” established by consensus committees can lead to a “false gold standard” [18] and could potentially lead to erroneous conclusions. In addition, previous studies have used image sets compressed by using a fixed quality factor for all images. Since the actual compression ratio for a given quality factor is image dependent, compressing a set of image sequences with a fixed quality factor will result in a large standard deviation in compression ratios (e.g. for 100 images at 20:1 one obtains a standard deviation of 4.2). Unlike other studies [8], in the present study a fixed compression ratio was achieved for all images, making our evaluation image compression more precise. In addition, the use of simulated signals does not present the problem of establishing a gold standard. Still, our study agrees with previous studies that found that 15:1 image compression does not significantly degrade detectability of the studied morphologic feature. The current study generalizes this finding to the feature-stabilized displays.
5. Conclusions
With the advent of digital Cathlabs, the use of feature-stabilized and 15:1 compressed coronary angiograms may allow reduced storage limitations and transmission time while improving human detection of morphological features over conventional uncompressed display methods.
6. Appendix
The imaging process is modeled as consisting of an absorption stage and a scattering stage. The incident x-rays are absorbed by the patient body and with higher attenuation by the contrast filled arteries. The attenuated x-rays are then either transmitted directly to the detector as the primary component of the radiation or are scattered to the detector.
In this model the image I(x,y) is composed of a primary component due to the primary beam, P(x,y) and a secondary due to scattered rays, S(x,y):
The primary component of the image is related to the incident rays (Io) and the densities of the projected patient (Dp)and contrast filled artery( Da):
We wish to insert a simulated object in the scene so that the primary component of the image needs to attenuated by the density of the simulated object (e-Ds(x,y)):
where Ps(x,y) is the primary including the simulated object. The density of the simulated object is given by Ds = μts(x,y) where m is the attenuation coefficient and ts(x,y); is the thickness of the contrast filled simulated object traversed by the x-rays as a function of the image position. To embed a simulated object into a real angiographic image, we first subtract an estimate of the secondary from the image, then attenuate the remaining estimated primary corresponding to the simulated artery, and then add back the secondary component:
where Is(x,y) is the image after the simulated object is inserted into the image and I(x,y) is the original image prior to insertion of the simulated object. The secondary component S(x,y) is not known a priori but can be approximated based on previous work.
Love and Kruger [10] used imaged phantoms to empirically show that the secondary image component due to scattering and veiling glare can be approximated by the convolution of the image and a double exponential kernel [11]:
where α= 0.483, β= 7.69. The convolution kernel is given by:
where A = B = 2log(2)/√75
The reduction in the density (Eq. A. 4) of the artery due to the filling defect is given by:
where xo and yo are the locations of the center of the filling defect, r is the radius, and κ is a contrast factor.
7. Acknowledgements
The authors would like to thank Cedric Heath and George Ruan for their participation as observers in the study. This research was supported by NIH-HL RO1 53455.
References and links
1. S. E. Nissen et al., “Cardiac Aniography Without Cine Film: Erecting a Tower of Babel in the Cardiac Catheterization Laboratory,” J. Am. College Cardiology 24, 834–837 (1994). [CrossRef]
2. J. S. Whiting, D. Honig, D. Carterette, and N. L. Eigler, “Observer performance in dynamic displays: effect of frame rate on visual signal detection in noisy images,” Proc. SPIE 1453, 165–167 (1991). [CrossRef]
3. M. P. Eckstein, J. S. Whiting, and J. P. Thomas, “Detection and discrimination of moving signals in Gaussian uncorrelated noise,” Proc. SPIE 2712, 9–25 (1996). [CrossRef]
4. M. Tapiovaara, “Efficiency of low-contrast detail detectability in fluoroscopic imaging,” Med. Phys. 24, 655–64 (1997). [CrossRef] [PubMed]
5. D. L. Wilson, K. N. Jabri, P. Xue, and R. Aufrichtig, “Perceived noise versus display noise in temporally filtered image sequences,” J. Electron. Imaging 5, 490–495 (1996). [CrossRef]
6. R. Aufrichtig, C. W. Thomas CW, P. Xue, and D. L. Wilson, “Model for perception of pulsed fluoroscopy image sequences,” J. Opt. Soc. Am. A 11, 3167–3176 (1994). [CrossRef]
7. N. L. Eigler, M. P. Eckstein, K. Maher, D. Honig, and J. S. Whiting, “Effect of a stenosis stabilized display on morphological feature detection,” Circ , 89, 2700–2709 (1994). [CrossRef]
8. W. A. Baker et al., “Lossy (15:1) JPEG compression of digital coronary angiograms does not limit detection of subtle morphological features,” Circ. 96, 1157–1164 (1997). [CrossRef]
9. C. A. Morioka, M. P. Eckstein, J. L. Bartroff, J. Hauseleiter, and J. S. Whiting, “JPEG vs. Wavelet image compression of x-ray coronary angiograms,” Opt. Express (to be submitted).
10. L. A. Love and R A. Kruger, “Scatter estimation for a digital radiographic system using convolution filtering,” Med. Phys. 14, 178–185, (1987). [CrossRef] [PubMed]
11. Note that in this method the scattering and veiling glare is not corrected for the presence of the new simulated object.
12. D. M. Green and J. A. Swets, Signal Detection Theory and Psychophysics (Wiley, N.Y.1966).
13. A. E. Burgess and H. Ghandeharian, “Visual signal detection. II. Signal-location identification,” J. Opt. Soc. Am. A 1, 906–910 (1984). [CrossRef] [PubMed]
14. M. P. Eckstein and J. S. Whiting, “Visual signal detection in structured backgrounds I. Effect of number of possible locations and signal contrast,” J. Opt. Soc. Am. A 13, 1777–1787 (1996). [CrossRef]
15. J. S. Whiting, M. P. Eckstein, C. A. Morioka, and N. L. Eigler, “Effect of additive noise, signal contrast and feature motion on signal detection in structured noise,” Proc. SPIE 2712, 39–47 (1996).
16. J. S. Whiting, M. P. Eckstein, and N. L Eigler, “Effect of motion on feature detectability in dynamicaly displayed coronary arteries,” Fifth Far West Image Perception Conference, Newport , RI (1993).
17. R. A. Close and J. S. Whiting, “Decomposition of projection image sequences into moving layers,” in Computer Assisted Radiology and Surgery, Editors, H. U. Lemke, M. W. Vannier, K. Inamura, and A. Farman, Elsevier Science , 143-146 (1998).
18. M. P. Eckstein, T. D. Wickens, G. Aharonov, G. Ruan, C. A. Morioka, and J. S. Whiting, “Quantifying the limitations of the use of consensus expert committees in ROC studies,” Proc. SPIE 3340, 128–134 (1998). [CrossRef]
