Myocardial Isolation and Sampling

Newer quantitative methods have been developed to overcome the major limitations described. These methods use several automatic image-identification techniques for isolation of the left myocardium from the remainder of the image.¹⁴ Once the left myocardium is identified, the apex and base, the coordinates of the central axis of the ventricular chamber, and a limiting radius for count profile search are determined automatically. In the majority of cases, operator interaction is required only for verification of automatically determined parameters. If at any time these automated programs fail to locate any of the features, they will branch to an interactive mode and require the operator to select the parameters manually.

Once the LV has been isolated from the myocardial perfusion scan, these automated programs extract the 3D LV myocardial count distribution. This sampling is done using either a hybrid, two-part, 3D sampling scheme of stacked short-axis slices¹⁵ or an ellipsoidal sampling of the 3D distribution inside the ellipsoid.¹⁶ In the hybrid sampling approach used in the CEqual program, the apical region of the myocardium is sampled using spherical coordinates, and the rest of the myocardium is sampled using cylindrical coordinates. This approach promotes a radial sampling that is mostly perpendicular to the myocardial wall for all points and thus results in a more accurate representation of the perfusion distribution with minimal sampling artifacts. Following operator verification of the automatically derived features, the 3D maximum count myocardial distribution is extracted from all stacked short-axis tomograms.¹⁶ During cylindrical sampling, maximum count circumferential profiles, each composed of 40 points, are automatically generated from the short-axis slices. During spherical sampling, maximal counts are extracted from the apical region, with the number of samples proportional to the apical area from which they are extracted. The combination of the cylindrical and spherical coordinates forms a set of 3D count profiles representing the myocardial tracer uptake. In the approach used by the QPS program, the sampling geometry is performed using an ellipsoidal rather than a hybrid model, and count profiles from endocardium to epicardium are used rather than maximal counts. Other methods used variations of these approaches, including a slice-by-slice maximal count circumferential profile extraction.^17,18

These 3D count distributions are generated for the stress and the rest myocardial perfusion distributions. A normalized percent change between stress and rest is also calculated as a reversibility distribution.¹⁹ The most normal region of the stress distribution is used for normalizing the rest to the stress distribution.

Normal Databases and Criteria for Abnormality

Once the stress and rest count distributions have been extracted from a patient’s LV myocardium and the reversibility distribution determined, they are compared to normal patterns found in computerized databases. These normal databases are usually generated from patients with a low likelihood of CAD. Although there is variation among methods as to how best to define these patients, in general, patients with less than 5% probability of CAD are used. This probability is based on the Diamond and Forester²⁰ criteria that uses a Bayesian analysis of the patient’s age, sex, symptoms, and the results of other noninvasive tests such as stress EKG.

The stress, rest, and reversibility distributions from males and females are separately combined to produce gender-matched normal files. This is done to help account for differences in body habitus that cause changes in the expected normal patterns between men and women due to photon attenuation. The distributions from these normal patients are statistically combined to provide the mean normal regional LV normalized count distribution and its corresponding regional standard deviation for the stress, rest, and reversibility distributions.^11,21

The next step in this process is to determine the criteria for abnormality. These criteria are usually determined as the number of standard deviations below the mean for each region in the stress and rest distributions and above the mean for the reversibility distribution. These numbers then represent the cutoff points between a normal and an abnormal region. Here also there is variation between methods. Some methods always use two standard deviations below the mean, some always use 2.5, some allow it to vary from patient to patient by the interpreter, and others determine the optimal cutoff criteria.

Determination of the optimal cutoff criteria for each region requires that perfusion studies from both normal and abnormal patients be used. In the normal databases used by the CEqual processing program, for example, the optimal cutoff point is determined using receiver operator curve (ROC) analysis.²¹ By varying the standard deviation (and/or any other quantitative parameter used, such as the extent of the abnormality) the determination of “normal” or “abnormal” by the program is compared to a gold standard such as expert visual analysis or catheterization results. ROC curves are then generated by plotting the true-positive rate versus the false-positive rate. The best cutoff is then determined as the desired tradeoff between the expected sensitivity and specificity for the detection and localization of CAD. Depending on the developers’ strategy, optimal criteria could mean the best tradeoff between sensitivity and specificity, or high-sensitivity with lower specificity, or higher specificity with lower sensitivity. These results can and do vary between methods, but they can also vary between normal databases used by the same program. It is imperative that the physician using these programs has reviewed published validations performed against prospective populations to understand the true performance characteristics of the quantitative software/normal database combination they are using.

Quantitative Parameters

The programs for quantifying LV myocardial perfusion report findings in terms of quantitative parameters. One typical parameter used is the extent of the perfusion abnormality. This extent is expressed either as a percentage of the total LV epicardial area that is hypoperfused or, in systems that calculate total LV mass, as the mass in grams that is hypoperfused. Usually in these systems, the extent of the hypoperfused region that reverses is also reported.

Quantitative parameters that take into account both the extent and severity of the abnormality are also determined. The total severity score is calculated as the total number of standard deviations below the mean normal distribution for regions exhibiting perfusion defects.²² The sum stress score is the sum of segmental scores in a (17 or 20) segmental model of the LV where each segment is scored from 0 (normal) to 4 (perfusion absent).²³ These scores are either manually or automatically assigned. Automatic assignment is performed by comparison to the normal database, with the scores calibrated to interpretations by human experts.

It is important to understand the thresholds that are used for differentiating normal from hypoperfused regions and how these thresholds are used to validate the various commercial programs. This knowledge will prevent the interpreter from misinterpreting small statistical fluctuations as abnormal hypoperfusion. For example, in the Emory Cardiac Toolbox (ECTb) program that uses the CEqual algorithm for quantification of myocardial perfusion, the key quantitative parameter used for identifying abnormality is the extent of the abnormality expressed as percent of the abnormal area. With this technique, at least 3% of the entire LV myocardium has to be hypoperfused before the patient is deemed to be abnormal. Similarly, at least 10% of the left anterior descending (LAD) region, 10% of the left circumflex (LCx) region, and 12% of the right coronary artery (RCA) region must be hypoperfused before CAD is localized to that vascular territory. Similarly, the extent of the stress perfusion defect has to reverse (improve) by at least 5% at rest for the region to be deemed partially reversible and at least 15% for it to be considered to improve post revascularization.

Commercial Implementations

Variations of the methods described thus for have been commercialized in at least eight separate products at the time of this writing. Three of these methods have become the most popular. These are the ECTb developed at Emory University,^24,25 Corridor 4DM developed at the University of Michigan,^26,27 and QPS/QGS developed at Cedars-Sinai in Los Angeles.^16,28 A fourth method also used commercially is the Wackers-Liu CQ software developed at Yale University.^17,29

These software packages have become popular because (1) they are automated; (2) they (usually) integrate image display, perfusion quantification, and functional quantification in one package; and (3) they are well validated.^25,30,31 Examples of applying three commercial programs to quantify the same patient are shown in Figures 4-3 through 4-5. It is important to understand that the implementation of these programs varies from vendor to vendor. Thus, even though one should expect that the same program will yield the same results given the same perfusion study, the ease and speed of obtaining and displaying results may vary not only between the packages but also from vendor to vendor and between versions of the same program.

Figure 4-3 Perfusion quantification results obtained from processing a patient with ECTb. The top row shows the quantitated stress (left) and rest (middle) perfusion and the difference between the two (right). The middle row shows the stress blackout map (left), the rest blackout map (middle), and the whiteout reversibility map (right). The bottom left map displays the number of standard deviations (SD) below normal of each myocardial sample of the stress (left) and rest (middle) perfusion, and the number of SDs above normal for reversibility (right). This patient had a completely reversible anteroseptal perfusion defect.

Figure 4-4 Perfusion quantification results obtained from processing the same patient as in Figure 4-3 with QPS. The left side of the figure (A) shows slices from the patient; the top three rows contain short-axis slices, and the bottom two rows show horizontal long-axis and vertical long-axis slices, respectively. The right side of the figure (B) shows the results of quantitation: the stress blackout map (top), the rest blackout map (middle), and the whiteout reversibility map (bottom). Note the similarities of these results compared with those of Figure 4-3.

Figure 4-5 Perfusion quantification results obtained from processing the same patient as in Figure 4-3 with Corridor 4DM. The top row shows the quantitated stress (left) and stress blackout map (right). The middle row shows the quantitated rest (left) and rest blackout map (right). The bottom row shows the difference between rest and stress perfusion (left) along with the whiteout reversibility map (right). Note the similarities of these results with those of Figures 4-3 and 4-4.

To date, no definitive study has been performed to compare the relative ease of operation, robustness, and accuracy of these programs. It is well accepted that in general the automation in these programs is quite robust, and they yield similar diagnostic accuracy.

New Databases

As new acquisition protocols and/or new perfusion tracers become popular, new normal databases must be developed. For example, for ECTb, gender-matched normal databases have been defined²¹ and validated for the following SPECT protocols:

All protocols used treadmill exercise to stress the patients.

The choice of which radiopharmaceutical and/or protocol should be used is more of a clinical question or a question of laboratory logistics and is beyond the scope of this chapter. However, there are a number of questions related to these normal databases that are very often asked and merit further discussion.

One concern is whether the normal databases developed using patients stressed with exercise can be used for patients undergoing pharmacologic stress. It is evident by looking at these studies that patients imaged after pharmacologic stress have more background activity and more myocardial activity than patients undergoing treadmill exercise. Nevertheless, the relative distributions in normal and CAD patients are similar enough that when the same normal database is used for both forms of stress, it results in similar diagnostic accuracy.³³ Although it would be ideal to have separate databases for protocols using pharmacologic stress, the development cost would be prohibitive.

A second concern is whether the normal database developed for ²⁰¹Tl-stress/redistribution studies may be used for stress/reinjection protocols. The stress protocol is the same for both studies, so no new errors should be introduced. However, the reinjection images do appear different from the redistribution images. Nevertheless, in the CEqual program used by ECTb, the resting distribution is not quantified, but rather the reversibility or change between rest and stress is analyzed. Since reinjection is supposed to result in a more marked difference between the two physiologic states, it is actually easier for the program to detect this difference in the form of defect reversibility. Thus, the stress/redistribution ²⁰¹Tl-normal database may be used to quantify stress/reinjection studies.

The last concern is whether the normal database for the ^99mTc-sestamibi protocols may be used for ^99mTc- tetrofosmin studies and vice versa. Although there are some subtle differences in how these two radiopharmaceuticals are distributed in the body, the main parameters driving the final count distributions in the images are the type of collimators and filters used for a given count distribution. It appears that these normal databases are interchangeable, but additional studies are required to confirm these observations.

Polar Maps

Polar maps, or bull’s-eye displays, are the standard for viewing circumferential profiles. They allow a quick and comprehensive overview of the circumferential samples from all slices by combining these into a color-coded image. The points of each circumferential profile are assigned a color based on normalized count values, and the colored profiles are shaped into concentric rings. The most apical slice processed with circumferential profiles forms the center of the polar map, and each successive profile from each successive short-axis slice is displayed as a new ring surrounding the previous. The most basal slice of the LV makes up the outermost ring of the polar map. Figure 4-3 shows polar maps created from applying the CEqual quantification method to a perfusion study.

The use of color can help identify abnormal areas at a glance as well. Abnormal regions from the stress study can be assigned a black color, thus creating a blackout map. Blacked-out areas that normalize at rest are color-coded white, thus creating a whiteout reversibility map.¹⁹ This can also be seen in Figure 4-3. Additional maps—for example, a standard deviation map that shows the number of standard deviations below normal of each point in each circumferential profile—can aid in evaluation of the study by indicating the severity of any abnormality.

Polar maps, while offering a comprehensive view of the quantitation results, distort the size and shape of the myocardium and any defects. There have been numerous improvements in the basic polar map display to help overcome some of these problems.¹⁵ For instance, “distance-weighted” maps are created so that each ring is the same thickness. These maps have been shown to be useful for accurate localization of abnormalities. “Volume-weighted” maps are constructed such that the area of each ring is proportional to the volume of the corresponding slice. This type of map has been shown to be best for estimating defect size. However, more realistic displays have been introduced that do not suffer from the distortions of polar maps.

Three-Dimensional Displays

Three-dimensional graphics techniques can be used to overlay results of perfusion quantification onto a representation of a specific patient’s LV. Figure 4-6 displays the same information seen in the polar maps of Figure 4-3, using a 3D representation. In its most basic form, the pixel locations of the maximal-count myocardial points sampled during quantitation are used to estimate the myocardial surface. More sophisticated methods may detect the epicardial surface boundaries of the perfusion scan.^16,24,34 These points can be connected into triangles, which are then color-coded similarly to the polar map; details of how such displays can be created may be found in the book by Watt, for example.³⁵ Such displays can routinely be rotated in real time and viewed from any angle with current computer power. They have the advantage of showing the actual size and shape of the LV and the extent and location of any defect in a very realistic manner. Some studies have shown that the 3D models are more accurate for evaluating the size and location of perfusion defects than polar maps³⁶ or slice-by-slice displays.³⁷

Figure 4-6 Three-dimensional surface-rendered displays of stress blackout and reversibility whiteout information from the same patient as in Figure 4-3. From the top, clockwise, are shown an anterior, lateral, inferior, and septal view. An apical view is shown in the middle of the display.

Stay updated, free articles. Join our Telegram channel

Join