Published in: Medical Imaging 1995: Physiology and Function from Multidimensional Images, Eric A. Hoffman, Editor, Proc. SPIE 2433, pages 26-36 (1995).

Perfusion Deficit Versus Anatomic Visualization in Detection of Pulmonary Emboli via Electron-Beam CT: Validation in Swine

E. A. Hoffman, J. K. Tajik, G. Petersen, T. J. Reiners, B. H. Thompson, W. Stanford

Departments of Radiology and Physiology
University of Iowa College of Medicine
Iowa City, IA. 52242

Table of Contents

ABSTRACT

INTRODUCTION

METHODS

RESULTS

CONCLUSION

REFERENCES

ACKNOWLEDGEMENTS

With the goal of improving the accuracy of pulmonary emboli diagnosis, investigators have recently taken advantage of high speed X-ray CT scanning accomplished via use of Imatron's electron beam CT (EBCT) scanner 2. Data from human studies 3,4 indicate that, while CT was able to pick up what the investigators deemed to be "clinically significant emboli," the CT visualization of emboli relative to the "gold standard" of angiography showed a sensitivity of only 65%. Stanford has previously demonstrated considerable accuracy in the use of CT to directly visualize contrast enhanced gel foam pulmonary emboli in dogs5 and autologous clots in pigs6. The contrast enhanced gel foam emboli study is probably not pertinent to an evaluation of native emboli detection, and the autologous clots in the pig studies reported previously were presumably of a size and stability such that they most likely represented emboli similar to the "clinically more significant" clots detected in the human studies3,4.

In this study, we link together an evaluation of pulmonary emboli based upon direct visualization via thin sliced CT with an accompanying study integrating a non-contrast thin slice volumetric scan with a high temporal resolution blood flow study using the high speed scanning capabilities of the EBCT scanner.

Scan Protocol 1:

60-3mm contiguous slices were acquired during 1.5cc / sec infusion of contrast media (Hypaque-76). The scan aperture was 100msec. Images were not ECG gated and the ventilator remained on during scanning with the notion that patients scanned under similar protocols would not be able to remain apneic during scanning. The EBCT scanner allows for acquisition of a maximum of 40 sections in a single scan set-up procedure. Thus, at the end of 40 slices, contrast infusion was halted, the scanner was re-set to continue gathering the remaining 20 slices, and infusion of contrast and scanning continued. Slice locations were determined via use of an anterior-posterior and a lateral scout scan and sections were positioned such that the last slice, in an apex to base scanning direction, was located to include the most caudal lung region. Most often this meant that the apical lung regions were not scanned.

.

Scan Protocol 2:

40 3mm contiguous slices (100 msec scan aperture) were gathered with no contrast infusion during apnea with the lungs held at functional residual capacity. Slice acquisition was gated to the ECG and slices were acquired at the QRS complex. The cephalad-caudal extent of scanning was selected to cover the carina to the base of the lung. Because experience taught us that the most likely regions for emboli locations were in the caudal half of the lung, for purposes of protocol brevity, we only scanned 40 sections (slice limit of the EBCT scanner for one acquisition set-up). In a patient setting, if it is important to evaluate the full lung, one would have to scan the lung in several acquisitions under the current scanner configuration. This limitation will likely be eliminated in future versions of the EBCT scanner. This data set was gathered such that, once regions of flow deficit were located from scans acquired under protocol 3, vessels feeding the flow deficit regions could be traced back to their parent branch and thus the likely location of the embolus would be identified without the contrast load given in Protocol 1.

Scan Protocol 3:

This protocol was used to follow the passage of a sharp bolus injection of contrast agent as it passed through the lung parenchyma for the purposes of assessing indices of regional pulmonary blood flow (parenchymal perfusion). The protocol which we use for pulmonary blood flow evaluation is described only briefly here. Greater detail is given in Hoffman et al.7 High temporal resolution images were generated by sequentially sweeping each of the four target rings on the EBCT scanner at selected points in time (gated to the subject's ECG). Each sweep of a target ring requires 50 msec followed by a 8 msec reset time, requiring a total of only 224 msec for all four targets. A collimator is used so that the x-ray beam from each target sweep spans approximately 16 mm of the body and focuses on the two detector rings, generating a pair of 8mm thick cross sections. Contrast was injected into the right side of the heart. Scans were taken at multiple time points to sample the "time-intensity" curves. Due to the 224 msec scan time (4 pairs of 50 msec scans with 8msec interscan delays), all eight slices are not acquired at an identical point in the cardiac cycle. Since this delay remains constant, gating scans to the ECG insures that each slice is always sampled at the same location in the cardiac cycle over multiple heart beats (i.e slice pair 3 and 4 will always be at peak QRS plus 58 msec over multiple cardiac cycles). Scans at multiple time points were acquired during a single breath hold, insuring that lung volume remained constant during the scan period. Contrast was injected at a rate of 10cc/sec over 2 sec and scanning commenced one heart beat prior to contrast injection. Scanning occurred at every heart beat for 6 heart beats to assure acquisition of the peak of the contrast curve, and then scanning occurred at alternate heart beats for then next 4 heart beats to maximize the amount of the tail of the curve sampled. Again, there was a slice limit in this mode of EBCT scanning. A maximum of 80 slices could be acquired in a single acquisition. In this mode of scanning, with 4 pairs of 8mm thick sections and 4mm gaps between contiguous pairs of slices, 7.6cm of the lung is spanned which is less than the 12 cm spanned via the 40 3mm thick sections of Protocol 2. We chose to locate these 8 high temporal resolution sections so as to exclude the most basal portion of the lung in the extreme costophrenic angle just above where the diaphragm and rib cage become juxtaposed.

Following final scanning the pigs were euthanized with pentobarbital, frozen, and the thorax sectioned into 10mm cuts for evaluation of clot location by a pathologist (based upon location of suture imbedded in injected clot).

Using data from scan protocol 3, we evaluated the distribution of pulmonary blood flow in the pig before and after the clot injection. Scan protocol 2 allowed us to link the perfusion data with high resolution volumetric image data sets for determination of the anatomic distribution of vessels feeding the region of flow deficit. Image data from scan protocol 1 were reviewed by two radiologists in a double blinded protocol.

Figure 1: Time intensity curves taken from the right main pulmonary artery (left panel) and right lung parenchyma at mid chest level (right panel)

Figure 2: The above images are a montage taken from the computer screen depicting an image based perfusion analysis module of VIDA in which time intensity curves are used to evaluate tissue perfusion. In the case shown above, the peak of the parenchymal curve is compared with the area under the pulmonary arterial curve to evaluate a number of physiologic parameters including regional air content, tissue content, blood content, blood flow normalized to tissue content (ml/gm parenchyma/min) or blood flow normalized to air content, mean transit time, arrival time, etc. Criteria can be set (lower left panel) such that only the image data is accepted from pixels representing lung parenchyma by throwing out pixels in which contrast arrives early (representing an artery) or late (representing a vein), etc. In this particular data set, flow in region 2 represented normal parenchymal perfusion, and flow in region 1 was found to represent an unperfused region.

Figure 1 demonstrates a set of typical time intensity curves obtained for the pulmonary artery and peripheral lung parenchyma from scan data gathered under Protocol 3. From a rise of approximately 1,000 Hounsfield units in the right pulmonary artery, we see an associated rise of 115 hounsfield units in the lung parenchyma at mid chest level. These image data are used to calculate parameters of pulmonary blood flow along with parameters such as mean transit time, arrival time, regional blood volume, regional tissue volume, etc. As shown in Figure 2, we use a module of our software package, VIDA 8 to evaluate the parameters of pulmonary blood flow and to provide a color coded map of each parameter superimposed onto the corresponding location in the high spatial resolution scan provided by Protocol 2. Since the animals did not move between Protocol 2 and 3 and, because scanning was done at the same lung volume; knowledge of the scanner geometry in the two studies, we are able to provide the map of one data set into the other 8. The module depicted in Figure 2 has various algorithms to calculate flow parameters, each one customized to accept the time intensity information associated with particular injection schemes and/or assuming that the injection occurred at a particular injection site. Algorithms are tailored, in part, to be compatible with the unique physiology of a particular organ of interest. Additional algorithms can be easily added into the module so long as they fit the basic criteria that they use a single input and a single output function as the basis for their modeling of blood flow.

[More information regarding this module can be found on the internet by using a web browser (Mosaic or Netscape) and looking at the following URL: http://everest.radiology.uiowa.edu/home.html. Information can be found by looking at either the "current research" pages or the VIDA tutorials. In the VIDA tutorials look under "tools for analyzing and quantifying images" under the VIDA table of contents. Another area to look is under the NLM teleradiology project and then under application specific tutorials, find the one on "Lung blood flow measurements."]

From the evaluation of image data from Protocol 3, we found that the average mean transit time in all lung regions from one representative pig was 6.57 seconds +1.34 seconds (S.D.) and the mean arrival time relative to the first scan which occurred one heart beat prior to injection of contrast into the inferior vena cava was 2.67 seconds + 0.46 seconds (S.D.). While arrival time varied slightly from pig to pig depending upon the cardiac output etc, each pig showed a similarly homogeneous arrival time andmean transit time throughout the lung parenchyma. Thus, the maximum enhancement of the lung parenchyma occurred at a common time point everywhere in the lung for a given pig. The implication of this is that if the goal of the study is not to quantify the exact parameters of flow, but to simply answer the question as to whether or not flow is present, one does not need the power of the VIDA module represented in Figure 2. A few samples of the parenchymal curves as depicted in the right panel of Figure 1 provide an adequate description of the time point at which all lung parenchyma achieved maximal brightness with the passage of the bolus contrast agent.

Figure 3: Since the lung is held at a fixed volume during scanning, and data acquisition is gated to the ECG, cross sectional images at a given thoracic level gathered at each of multiple cardiac cycles are well aligned. By subtracting the slice data acquired at the first time point (one heart beat prior to contrast injection) from each of the other temporally sequenced slices scanned at the same thoracic level, one is left with an image of the pixel enhancement occurring due to the passage of the injected bolus contrast agent. Since we have demonstrated that arrival time and mean transit time is fairly uniform everywhere in the lung (within the course time frame of the sampling carried out under these studies), we can look at the subtraction image at the single time point where peak contrast enhancement occurred and assess whether or not there is a flow void or not as an index of the presence of a pulmonary embolus.

To aid in visualizing the relative distribution of pulmonary blood flow, as depicted in Figure 3, we digitally subtracted, pixel by pixel, the first time point at each slice level from every other time point at that level. Through visualization of the subtraction image at the time of peak contrast enhancement, we were able to identify those parenchymal regions receiving blood flow and to distinguish those parenchymal regions devoid of flow due the presence of a pulmonary emboli in the parent pulmonary arterial branch feeding that region of lung. From the use of these subtraction images, we evaluated whether or not there was evidence of pulmonary emboli from the data gathered under Protocol 3. Observation of a flow deficit observed in Protocol 3 was combined with use of the scan data from Protocol 2 to identify the parent pulmonary arterial branch where the pulmonary emboli was likely located. This location was compared with the blinded observations made by the two Radiologists reviewing data gathered under Protocol 1.

Figure 4: This image is one from a pig scanned under Protocol 1. In this protocol, a slow drip delivery of contrast was used while imaging 3mm thin sections of the thorax for the purposes of visualizing the pulmonary emboli as a flow void (unenhanced luminal region) within the pulmonary artery. As shown by the arrow, a pulmonary embolus was detected for this pig in the vicinity of the left lower lobe.

Figure 5: Shown here is a photographic image of the frozen section in which a clot was found and verified based upon the presence of the suture material which was passed through the clot at the time of injection. This clot corresponds to the clot detected and displayed in Figure 4.

Figure 6: This sequence of images represents the eight sections scanned under Protocol 3 for the same pig depicted in Figures 4 and 5. Sectioning is from mid-chest to lung base going from top left to bottom right. Images are derived from the digital subtraction of the first time point from the time point where maximal parenchymal enhancement was found. These images represent parenchymal flow distribution prior to clot injection

Figure 7: This is the equivalent image set as shown in figure 6, from the same pig depicted in figures 4-6. Data represents the distribution of pulmonary blood flow after clot injection. Two flow voids were found, one in the right lower lobe and one in the left lower lobe as marked by the arrows. Note that these flow voids have quite distinct border characteristic of the flow voids observed to be associated with pulmonary emboli. The flow void shown on the right of each image (left lung since images are viewed from the caudal surface looking towards the lung apex) in the bottom row represents the flow void associated with the embolus observed from protocol 1 and depicted in figures 4 and 5. The flow void shown on the left of each image in the lower row represent a pulmonary embolus in the right lower lobe not found by either protocol 1, or via post mortem direct visualization. This flow void is presumably associated with a pulmonary emboli caused by a shattering of the injected clot, and thus post mortem evaluation did not pick up the clot because the suture material traveled with that portion of the clot which went to the left lung.

The image shown in Figure 4 demonstrates the presence of a pulmonary emboli found in the evaluation of image data from Protocol 1, and the image shown in Figure 5 demonstrates the post mortem verification of the presence of a clot in the region identified from the image shown in Figure 4. In Figures 6 and 7, we demonstrate the subtraction images at the time point of peak contrast enhancement observed before and after (respectively) injection of the autologous clot. These data are from the same pig depicted in Figures 4 and 5. Note that in Figure 6 there is no evidence of a flow void, while in Figure 7 there are two flow voids marked by the two arrows in the lower row. As is standard, images are viewed from base to apex, and thus the left lung is on the viewer's right, and the dorsal lung surface is at the bottom and the ventral surface is at the top of each image. The flow void found in the left lower lobe corresponded well with the location of the embolus visualized in Figure 4, depicting image data from Protocol 1. However, the flow void found in the right lung and seen in Figure 7 but not Figure 6 (pre clot injection) was not found in image data of the same pig scanned under Protocol 1. This flow void in the right lung under Protocol 3 is most likely a pulmonary emboli derived from a fractionation of the injected clot as the clot passed through the right atria and ventricle. The portion of the clot which did not have the suture associated with it could not be identified in the post mortem evaluation, because the post mortem evaluation distinguished the injected clot specifically on the basis of the suture being present. Note that the edges of the flow voids shown in Figure 7 are distinct. This sharp edge was observed in all of the flow voids associated with pulmonary emboli in this study. Whenever there was a flow void present which had poorly defined edges, the flow void was found both before and after injection of the clot and was associated with regions of pulmonary edema. (with pulmonary edema, it is well documented that vessels are constricted locally so as to shunt blood to regions receiving better oxygenation).

In all cases, when functional deficits were found they were present both before and after injection of the clot or only after injection of the clot. When found before injection of the clot as well as after, these regions correlated with areas of pneumonia diagnosed via visual inspection of the CT slices. Comparison of pathology findings with functional image based assessment of emboli showed agreement in 13 out of 14 studies. In one case, a clot was located by pathology which was not found by functional imaging because scanning did not include the region of the embolus. Of considerable interest was the finding that in addition to locating the position of the primary clot found by pathology, functional imaging identified an additional 8 regions which presumably represented areas of flow deficit caused by clot fragmentation. The double blinded radiologic review of scans from protocol one found 5 of the clots identified by functional imaging and missed 17. Of the 17 missed, seven of these were locations also missed by the pathologic evaluation and thus representing clot fragments other than the fragment with the suture. In one study, the post clot scan for functional information was not performed because of premature death of the pig. 17 pigs scanned more recently have yet to be evaluated.

2. Boyd, D.P., and M.J. Lipton, "Cardiac computed tomography," Proceedings of the IEEE, Vol.71, pp.298-307, 1983.

3. Teigen, C.L., T.P. Maus, P.F. Sheedy II, C.M. Johnson, A.W. Stanson, and T.J. Welch, "Pulmonary embolism: diagnosis with electron-beam CT," Radiology, Vol. 188, pp. 839-845, 1993.

4. Teigen, C.L., T.P. Maus, P.F. Sheedy II, A.W. Stanson, C.M. Johnson, J.F. Breen, M.A. McKusick, "Pulmonary embolism: Diagnosis with contrast-enhanced electron-beam CT and comparison with pulmonary angiography," Radiology, Vol. 194, pp. 313-319, 1995.

5. Geraghty, J.J., W. Stanford, S.K. Landas, and J.R. Galvin, "Ultrafast computed tomography in experimental pulmonary embolism," Invest Radiol, Vol 27, pp. 60-63, 1992.

6. Stanford, W., T.J. Reiners, B.H. Thompson, S.K. Landas, and J.R. Galvin, "Contrast-enhanced thin slice ultrafast computed tomography for the detection of small pulmonary emboli. Studies using autologous emboli in the pig," Invest Radiol., Vol 29, pp. 184-187, 1994

7. Hoffman, E.A., J.K. Tajik, and S.D. Kugelmass, "Matching pulmonary structure and perfusion via combined dynamic multislice CT and thin-slice high-resolution CT," Computerized Medical Imaging and Graphics, Vol 19, pp. 101-112, 1995.

8. Hoffman, E.A., D. Gnanaprakasam, K.B. Gupta, J.D. Hoford, S.D. Kugelmass, and R.S. Kulawiec, "VIDA: An environment for multidimensional image display and analysis," SPIE Proceedings, Vol. 1660, pp. 694-711, 1992.

Reprint requests to:

Dr. Eric Hoffman, Dept. of Radiology

Univ of Iowa College of Med

200 Hawkins Drive, Iowa City, Iowa 52242