Medical Imaging
Computerized
Pergamon
and Graphics, Vol. 19, No. I, pp. 101-I 12, 1995 Copyright 0 1995 Elsevier Science Ltd Printed in the USA. All rights reserved 0895-61 I l/95 $9.50 + .OO
08956111(94)00035-2
MATCHING PULMONARY STRUCTURE AND PERFUSION VIA COMBINED DYNAMIC MULTISLICE CT AND THIN-SLICE HIGH-RESOLUTION CT Eric A. Hoffman,’ Jehangir K. Tajik,’ and Steven D. Kugelmass2*3 Cardiothoracic Imaging Research Center, Department of Radiology, University of Pennsylvania School of Medicine, Philadelphia, PA 19104, USA (Received
9 May 1994)
Abstract-Taking advantage of two scan modes of an electron beam CT scanner (lmatron), we have developed a method utilizing x-ray CT for relating pulmonary perfusion to global and regional anatomy. A high temporal resolution mode, used to follow bolus contrast agent, is combined with a high spatial resolution mode to obtain the structure-function fusion. A software module has been developed for our image analysis package (VIDA)@ to automatically calculate physiologic parameters of flow and integrate these color coded functional measurements into a corresponding high spatial resolution data set. We present the scanning methodology details and give examples from our physiologic based research to demonstrate strengths of combining dynamic and high resolution CT to uniquely characterize pulmonary normal and pathophysiology. Key
Words:Pulmonary Pulmonary
imaging, function
Perfusion,
Blood flow, High resolution
INTRODUCTION
CT, Dynamic
CT, Quantitative
imaging,
cardiac size and body posture dependent shifts in mediastinal mass, and regional pulmonary blood volume. These limitations have placed particularly large obstacles in the move to understand disease processes often related to disruptions of the integrated functioning of multiple organ systems. Our understanding of the determinants of ventilation and perfusion distribution is of paramount importance in, for instance, the understanding of: (a) the mechanisms of altered blood gasses during anesthesia; (b) the selection criteria for optimal body orientation during positive end expiratory pressure (PEEP) ventilation of patients with adult respiratory distress syndrome (ARDS); (c) the patchy distribution of reperfusion injury following pulmonary embolization; (d) the altered V/Q matching during exposure to high accelerative forces; (e) the possible effects of exposure to zero gravity during space travel; and to get quite esoteric; and (f) the possible mechanisms whereby animals such as the elephant, whale, or even the horse perhaps minimize V/Q inequalities by avoiding the supine body posture. Clinically, possibly the most common motivation for the evaluation of regional ventilation-perfusion relationships is to assess the presence of pulmonary emboli. Clinical quantitation of V/Q relationships along with investigations into the physiologic determinants of regional differences in pulmonary perfusion and
The primary role of the lung is to deliver ambient air (ventilation) and the body blood supply (perfusion) to an interface (the alveolar walls) for the exchange of carbon dioxide and oxygen. In the best case, regional ventilation and perfusion are matched to each other so that the ventilation-perfusion ratio (V/Q) is uniform everywhere. It is known that regional blood flow differences are ini part a function of the differences in arterial, venous and alveolar pressure, that hydrostatic pressure gradients play a role in determining regional blood volume, and regional blood flow can be influenced by neural-humoral mechanisms responding in part to regional oxygen tensions. Investigations to date have been in large part limited by the inability to correlate ventilation and perfusion parameters simultaneously with in vivo assessment of related anatomic detail such as rib cage and diaphragm geometry, pulmonary vascular branching patterns, mediastinal anatomy including
’ Current address: Division of Physiologic Imaging, Deptiment of Radiology, University of Iowa College of Medicine, Iowa _ City, IA 52242 I%A. * Current address: Bear Stems and Co., Inc., 245 Park Avenue, New York, NY 10167. ’ Correspondence should be addressed to Eric Hoffman, Ph.D., Division of Physiologic Imaging, Department of Radiology, University of Iowa College of Medicine, Iowa City, IA 52242 USA. 101
102
Computerized
Medical
Imaging and Graphics
ventilation have spawned numerous methods over the years. For ventilation measurements, nitrogen washout (I), multiple inert gas elimination (2), and radioactive xenon gas inhalation (3) techniques have been used while radiolabeled microaggregated albumin or radioactive microspheres injection in conjunction with external gamma counters (4) or lung excision (5) have provided perfusion data. More recently, the rise of powerful imaging methods such as positron emission tomography (PET), single photon emission computed tomography (SPECT), magnetic resonance imaging (MRI) and x-ray computed tomography (CT) have provided new approaches for in vivo study of pulmonary ventilation and perfusion (6).
Imaging as a tool to study pulmonary anatomy and function Early measurements of lung structure and function were derived from a unique CT scanner known as the Dynamic Spatial Reconstructor (DSR) (7). Capable of acquiring up to 240 contiguous 0.9 mm cross-sections in one-sixtieth of a second and repeating this acquisition 60 times per second, the DSR required development of specialized software for the visualization and analysis of three and four dimensional data sets. Display techniques such as shaded surface display and maximum intensity projections (ray casting) (8) were used for the three-dimensional (3D) visualization of the lungs and vascular tree. Oblique sectioning of the volumetric data sets, by extracting slices perpendicular to the airway or vessel long axis, allowed for accurate measurements of cross sectional areas and branch angles of the pulmonary vascular tree (9, IO) and airways. Hoffman and colleagues demonstrated the capabilities of in vivo imaging of lung, by using the DSR to show that lung volume could be accurately measured to within 3% of the excised lung water displacement (11) and that lung air content could be measured to within 7% with relative (region by region) accuracy reaching 3% (12). More recently, conventional CT scanners have been used to acquire highly collimated, thin slice (1 mm) cross sections of the lung. Reconstruction of the raw projection data with a high spatial frequency algorithm provides enhanced edge depiction, thereby generating high resolution (HRCT) slices for visualizing anatomic detail (13). The advent of spiral CT (14) has allowed for the acquisition of high resolution volumetric image data sets, and techniques originally developed for the DSR are being applied to the display of the pulmonary arteries (15). While providing impressive anatomic information, this approach has been restricted to the area in which conventional radiology has resided
January_February/l995,
Volume
19, Number
I
ever since Roentgen’s original application in the 1890s that is, depiction of structure. Functional depiction of the lungs has traditionally been limited to the radio-nuclear branch of radiology. Here, research into relating blood flow to anatomy, when structures cannot truly be visualized (i.e. PET scanning or post mortem dicing), has recently taken the form of fractal analysis, whereby fractal dimensions are calculated to indirectly infer the underlying anatomical characteristics (16). Some investigators have attempted to match data sets acquired from multiple imaging modalities (I 7) such as MRI (anatomy) and PET (function), but in large part such matching has been limited to the study of the brain (18). When it comes to the lung, its cyclic motion is less repeatable than that of the heart, causing difficulties in gating, while its air content causes problems for MR imaging due to magnetic susceptibility effects at airwater interfaces (19). Electron beam CT (EBCT) (20) on the other hand, has provided the ability to generate both high temporal and high spatial resolution detail of pulmonary function and anatomy. Since both structure and function can be imaged on the same scanner, we simply take care to monitor important physiologic parameters such as air flow, airway pressure, and the ECG so that the lung is always at the same volume during slice acquisition and that the heart is at the same location in its cycle, thereby eliminating cardiogenic motion of the parenchyma (21). By knowing the exact scanner geometries during the high spatial and high temporal resolution acquisition modes, we can easily map function to structure. In this paper, we describe our approach to image data acquisition and the subsequent post processing associated with tine CT scanning for providing information relating airway and vascular anatomy to regional parenchymal blood flow and air content. IMAGE
ACQUISITION
AND
ANALYSIS
The Imatron Fastrac C-100 tine X-ray CT scanner, depicted in Fig. 1, provides a unique platform for simultaneously assessing pulmonary anatomy and function. The C-100 offers both high resolution and dynamic scanning modes while eliminating the need for any target or detector motion by replacing conventional x-ray tubes with electron beam technology. Multiple views are generated by magnetically steering a focused electron beam along one of four hemicylindrical target rings positioned around a subject. Opposite the four target rings are two stationary detectors each encompassing a 210” arc around the subject. Crystal photodiodes in the detector rings are used for recording transmitted x-ray intensity with one detector ring being double populated for high spatial resolution scanning.
Pulmonary structure and perfusion
l
E. A.
HOFFMAN, J. K. TAJIK and S. D. KUGELMASS
103
Fig. I. Cross sectional view of the Imatron scanner. An electron beam from the attached gun is magnetically tocused and steered along one of four hemi-cylindrical tungsten target rings. Transmitted X-ray intensity is detected by two rings of crystal photo-diodes opposite the target rings. Notice the electron beams deflected from each target ring converging on the same detector pair, resulting in slightly fanned, rather than parallel slices in the dynamic mode. High spatial resolution images are acquired by moving a collimator between the target and detector rings, narrowing and focusing the deflected beam on the double-populated detector.
High resolution
scanning for pulmonary anatomy Resolution CT images (UHRCT) are acquired by moving a collimator between the target and detector rings, thereby narrowing the deflected electron beam and focusing it on the doubly populated detector ring. In this thin slice mode, the current scanner configuration produces one 3 mm slice by a 100 msec double sweep of the target ring. (Imatron is, at the time of this writing, now shipping scanners with 1.5 mm thin slice capability.) Higher quality images are achieved by allowing for four 50 msec sweeps, yielding an effective scan aperture of 200 msec. Advancing the patient table after each 100 msec sweep generates up to 40 3 mm contiguous, parallel slices, providing a volume image from the apex to base of the lung in 52 s. Optional continuous slow infusion of contrast material (1 cc/s) during the scan period allows enhanced high detail visualization of the pulmonary vascular tree. When pulmonary emboli are present, the contrast enhanced, flowing blood will reconstruct with a higher Hounsfield unit than the clotted embolic region, allowing for visual distinction. It should be noted here that it is critically important to suspend respiration during scanning, and lung volume must be held at the same volume in both the high spatial and high temporal resolution modes. Lung volume is most repeatably held at functional residual capacity (FRC). Recent work by Kalender et al. (22) have demonstrated a method for achieving a repeatable lung volume through the use of a mouth piece and nose clip, keeping track of air flow Illtrafast
High
in and out of the mouth via a pneumotachometer. The subject is asked to inspire to total lung capacity (TLC), expire to residual volume (RV), and then lung volume is clamped at 50% of the vital capacity (VC = TLCRV) during scanning. After approximately 10 or 15 s scan time, the airflow to the mouth piece can be again opened, allowing the subject to take a few breaths before again occluding the airway at 50% of VC and resuming scanning to acquire a full three-dimensional, thin slice data set. Ideally, each slice is gated to the ECG to eliminate cardiogenic motion effects (21). Since 200 msec scan times are long relative to the motion of the heart during systole, scanning should be gated to occur during late diastole. Imatron provides several options when reconstructing images with a high spatial frequency algorithm for enhanced edge depiction. Choosing a transverse slice with 256 by 256 elements covering a 15 cm field of view gives each picture element (pixel) a dimension of 0.59 mm on a side. Pixel dimensions can be reduced by either decreasing the field of view or increasing the size of the reconstruction matrix to 384 by 384 or 5 12 by 5 12. With a pixel dimension of 0.59 mm and a slice thickness of 3 mm, each element in the 3D image (voxel) then represents 0.001 cm3 in the 256 by 256 matrix. Making measurements directly on the high resolution volume images with knowledge of the voxel dimension, provides a method for calculating such information as vessel diameter, cross sectional area, segment length, etc.
Computerized
Medical
Imaging
and Graphics
January-February/l
995, Volume
19, Number
1
Fig. 2. Demonstration of the relationship between the high spatial and high temporal resolution geometries. The image on the left is a shaded surface display of vascular and airway structure seen from a UHRCT data set. The image on the right is of HTRCT blood flow data imbedded into a coronal section of the corresponding UHRCT image set and graphically depicts the thicker 8 mm slice pairs with 4 mm inter-pair gaps.
Dynamic scanning for pulmonary function High Temporal Resolution CT images (HTRCT) are generated by sequentially sweeping each of the four target rings at selected points in time (gated to the subject’s ECG). Each sweep of a target ring requires 50 msec followed by an 8 msec reset time, requiring a total of only 224 msec for all four targets to be swept. A collimator is used so that the x-ray beam from each target sweep spans 16 mm of the body and focuses on the two detector rings, thereby generating a pair of 8 mm thick cross sections. Due to cost limitations, scanner geometry is such that x-ray beams from each of the four parallel target rings are focused on the same detector pair, resulting in slightly fanned cross sections with a 4 mm gap between each of the four slice pairs. Sweeping all four targets in conjunction with bolus contrast injection generates eight spatially distributed 8 mm slices, providing functional data spanning 7.6 cm of the subject’s body per time point. In this case, with a pixel dimension of 0.59 mm and a slice thickness of 8 mm, the voxel volume is 0.003 cm3. Figure 2 illustrates the relationship between the high spatial and high temporal resolution scan geometries. If contrast is injected into the right side of the heart, opacified blood is pumped through the pulmonary artery into the lung parenchyma. The flow of contrast media enhanced blood into and out of the parenchyma causes a change in the tissue radiodensity (Hounsfield number) over time. Scans are taken at multiple time points to sample the “time-intensity” curves, which are then used to calculate functional data (see below). Due to the 224 msec scan time, all eight spatially distributed slices are not acquired at an identical point in the cardiac cycle. Since this delay remains constant, gating scans to the ECG ensures 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). Moreover, scans at multiple time points are acquired during a single breath hold, ensuring that lung volume remains constant during the scan period. Again, it is critical to hold lung volume at the same level as used during the UHRCT volumetric sequence if the two data sets are to be matched appropriately. The current scanner configuration has a memory limitation of 80 slices before scanning must be halted to transfer images to backup media. Incorporating this limitation into our scanning protocol, to adequately sample contrast arrival and outflow, scans are taken every heart beat for the first five or six beats and every second or third heartbeat thereafter. If all eight spatially distributed slices are desired, up to 10 time points may be sampled. In order to provide at least one baseline point for blood flow calculations (see below), bolus injection must be timed in a manner such that contrast arrives at least one heart beat after initiation of scanning. Acquiring 6 slices at a time reduces the apex to base span of functional data but allows sampling at 13 time points (78 slices). In the studies reported here, we injected approximately 0.7 cc/kg contrast directly into the right ventricular outflow tract (RVOT) in a 2 s bolus via a 6F NIH multihole injection catheter to achieve good contrast agent/blood mixing while minimizing contrast dispersion during passage through the lung parenchyma. The amount of contrast agent used (in this case determined for our studies of 15-25 kg dogs with an RVOT injection site) is designed to limit the peak contrast enhancement of the main pulmonary artery such that recon-
Pulmonary structure and perfusion
l
E.
A. HOFFMAN, J.
strutted voxel values do not exceed approximately 800 Hounstield units. This has empirically been found to minimize scatter and beam hardening artifacts encountered when using the Imatron scanner in the HTRCT mode. It is important to achieve near maximal enhancement (i.e., enhancement to 800 Hounsfield units) to provide an adequate signal in the low flow regions in the lung periphery, particularly in the nondependent (upper-most) lung regions. For human studies, it may be des,irable to keep the injection site as peripheral as possible to minimize the invasiveness of the procedure, taking care not to over pressurize the peripheral vein with a sharp bolus injection. Clearly, there are tradeoffs between the ideal technique and the limitations of the clinical environment. Slice localization Care must be taken to ensure that at least one HTRCT slice contains a cross section of the main right and left branch of the pulmonary artery for use in blood flow calculations as outlined below. Scout scans which include an anterior-posterior and a lateral projection are used to select the desired apical to basal extent of scanning for the UHRCT studies. The UHRCT scans are in turn used to determine the desired spatial location of scanning for the HTRCT study. Analysis of tine CT images Blood flow and its parameters can be calculated by applying an appropriate blood flow model to samples of regional lung density changes and changes in the feeding vessel due to contrast passage. As described by Wolfkiel (23), assuming a sharp enough bolus, such that the amount of contrast enhanced blood leaving a region is minimal prior to full bolus arrival, blood how can be calculated using equations derived from conventional microsphere approaches to blood flow analysis. F,C,(f)
=
$ + F"C,(f).
where F, = flow to the tissue from main feeding vessel C, = incoming concentration of indicator F, = flow away from tissue C,, = outgoing concentration of indicator A = accumulated amount of indicator in tissue. The last term of this equation can be neglected by assuming the amount of contrast enhanced blood leaving a region is minimal prior to the full bolus arrival. Furthermore, by definition, the accumulated amount of indicator in a sample is simply the volume of the sample multiplied by the concentration of indicator in the feeding vessel (A
105
K. TANK and S. D. KUCELMASS
= Ci*v). Since CT images exhibit a linear relationship between image brightness (Hounsfield number) and amount of contrast material, the amount of contrast present (A) in a region can be found from the magnitude (pealbase) of the regional “time-intensity” curve. Thus, in terms of CT scanning:
where = The Hounsfield unit measured at the peak of the parenchymal dilution curve unit measured prior to the CT,,,, = The Hounsfield arrival of contrast into the region of interest CT,, = The Hounsfield unit of contrast agent in the feeding pulmonary artery VP = The volume of parenchyma present in the region of interest.
CT,,,
Blood flow measurements from dynamic CT images can, therefore, be easily calculated by relating a derived regional “time-intensity” curve (a gamma variate fit to the sampled regional time-intensity points) to the raw time-intensity curve found in the vessel supplying the tissue (i.e. main right or left pulmonary artery). To express blood flow per gram of parenchyma (VP), we assume the region of interest (ROI) sampled within the lung to be composed of air and “water” (i.e. blood + parenchyma) (12). Because of the disparate densities of these two components, we can use the CT gray scale values (Hounsfield number) of the images to calculate the fraction of each component in the ROI. The water fraction, for example, can be calculated by subtracting the CT value of pure air from the mean CT value of a ROI to give a number relating to the amount (density) of parenchyma + blood present in the selected region. Because of the linear relationship between Hounsfield number and density, comparing this value to the continuum ranging from pure water (Hounsfield number = 0) to pure air (Hounslield number = - 1024) yields the percentage of the ROI which is water (parenchyma + blood). (parenchyma +
blood)
=
CF
.
-_‘fF bhti
(3)
a,r
The air fraction of an ROI, then, is simply 1.O - “water” fraction. Furthermore, the fraction of blood present within the ROI can be computed by comparing the “timeintensity” curve of the ROI to the curve in the feeding vessel.
Computerized Medical Imaging and Graphics
JanuaryyFebruary11995. Volume 19. Number I
Fig. 3. Multiple views of canine pulmonary vasculature demonstrating structural detail from the UHRCT. The top row shows the lung rotating about the right to left body axis while the bottom row shows the lung rotating
about the head to foot axis. The data set represents an inverted brightest voxel display following interactive editing whereby a region growing algorithm was used to isolate the non pulmonary structures and multiply them by 15%.
Blood = ! s 0
(4) CT,,(r)dr
Subtracting this result from the water fraction leaves the percentage of the ROI which is lung parenchyma. Multiplying the percentage of parenchyma, the number of voxels in the ROI, and the voxel dimension finally yields the volume of parenchyma (V,,) in cm3 (or grams since parenchymal density is nearly equal to 1 g = cm’). CUSTOMIZED
SOFTWARE PROCESSING
FOR
IMAGE
Tools for the analysis and display of multidimensional image data sets are provided by an in-house developed image analysis package dubbed VIDA@ (24). VIDA runs on UNIX workstations and is written in the X window system to provide an easy to use graphical interface to the various modules it comprises. The normal sections, oblique sections, and volume render modules of VIDA are particularly useful for the visualization and display of pulmonary structure. The “Normal Sections” module can be used to display sagittal, coronal and transverse slices from a 3D volume of the lungs, while the “Oblique Sections” module allows the user to generate slices at any angle or orientation through the lung. We use the “Volume Render” module for the 3D display of data from the HRCT scan. As shown in Fig. 3, brightest voxel projec-
tions through the high resolution images allows detailed visualization of the pulmonary vascular tree from any orientation. The brightest voxel (or “maximum intensity projection”) images are shown here with the grey scale inverted. A module which we originally developed for evaluating the upper airway is applicable to the quantitation of vascular anatomy (24). Figure 4 depicts the various windows of the “Tube Geometry” module. Here, we find the centroid of non-branching vessel segments through an iterative process. The user pre-processes the image such that the voxels comprising the vessel segment of interest are of a unique gray scale relative to the rest of the image. Next, a voxel within the vessel segment at the beginning and end points are identified. The algorithm which identifies the centroids of the vessel strikes a line between these two points, calculates the plane perpendicular to this line, and identifies (based on gray scale) all voxels in the plane comprising the vessel and then calculates the average vessel voxel location. This voxel location coupled with the original two end point locations are used to generate two new line segments, and the process is repeated to generate four line segments, etc., up to the number of iterations specified by the user. In the end, all of the points identified through this process become the list of tube centroids. Oblique sections are calculated along the full extent of the vessel so as to always be perpendicular to the local long axis of the vessel and vessel cross sectional area is graphed. The user can point to any location along this graph to obtain a numerical output of the cross sectional area along with the full gray scale image obliquely cutting the vessel at that level along with an image cutting through the length
Pulmonary structure and perfusion
l
E. A. HOFFMAN, J. K. TANK
and S. D.
KUGELMASS
107
Fig. 4. Multiple windows of the tube geometry module for the evaluation of vascular or airway geometry (crosssectional areas and diameters) at inter-branch locations (see text for extended details). The middle row of panels shows the orthogonal brightest pixel projections from which the user specifies the region of vessel for evaluation. A vessel center line is calculated and oblique slices always perpendicular to the vessel segment local long axis are calculated. The true vessel cross-sectional areas, along with maximum anterior-posterior (AP) and lateral (lat) diameters, are then calculated from the oblique sections. These values are shown graphically and in tabular form. The user can “click” on the graph (shown by vertical line) and an additional line will appear on the three orthogonal projections showing the locations from which the measurements were derived. Simultaneously, an oblique section at the vessel location will be displayed showing the actual vessel cross section (arrow in left panel of middle row).
of the vessel, showing the location identified in the graph. Wood et al. have recently reported the development of an approach to following branching structures, keeping track of segment lengths, branch angles, segment cross sectional areas, and associated nomenclature which will allow for a full description of the branching structure geometry (25). Calculating relevant physiologic data from dynamic CT scans, however, has, in the past, been limited to the tedious and time consuming task of manually sampling regions within the images and then computing the desired information. We have, therefore, developed an additional VIDA module to automate the analysis of tine x-ray CT images, thereby rapidly providing such physiologic relevant data as regional blood flow (normalized to tissue content and to air content), regional tissue, blood and air contents, regional mean transit and arrival times, time to peak, raw blood flow, peak opacification, area under the regional time-intensity curve, and the chi squared value for the curve. This module allows for the highly flexible application of essentially any blood flow model which utilizes a single input and single output function.
The upper left window of Fig. 5 shows the main panel of our program. Using this panel, dynamic CT image sets are loaded into computer memory and displayed in the bottom half of the window. Controls on this panel allow the user to view any transverse slice of the image set at any time point in the scan. An additional control allows the user to select to the size of a region to be sampled, from a one pixel by one pixel box to a thirteen by thirteen pixel box. Shown in the figure is a dynamic tine CT data set evaluated using a sample size chosen to be a five pixel by five pixel box. “Pointing and clicking” on an area of the displayed image produces a graph of the time-intensity curve of that region (upper middle in Fig. 5). Note that areas of high blood flow (i.e., pulmonary artery) show large changes in CT value (upper curve) while regions of lower flow (i.e., parenchyma) show smaller changes over time (lower curve). After designating a curve for the pulmonary artery, all subsequent “clicks” use this curve to calculate blood flow, air content, blood content, tissue content, mean transit time, arrival time, etc.
108
Computerized
Medical
Imaging
and Graphics
January-February/l
995, Volume
19, Number
Fig. 5. Composite image of VIDA’s blood flow module. Using menus on the main panel (upper left window), dynamic and high resolution volumetric CT image sets are loaded into computer memory and displayed in the bottom half of the window. Other controls on this panel allow the user to view any transverse slice of the image set at any time point in the scan, select the sample size, and enable batch mode processing. A loaded tine CT data set with the sample size chosen to be a five pixel by five pixel box is shown. “Pointing and clicking” on an area of the displayed image produces a graph of the time-intensity curve of that region (upper middle). Notice that areas of high blood flow (i.e., pulmonary artery) show large changes in CT value (upper curve) while regions of lower flow (i.e., parenchyma) show smaller changes over time (lower curve). Calculated values are displayed in a text window for the user to view (center bottom).
for that region as described above. These calculated values are displayed in a text window for the user to view (center bottom, Fig. 5). The real power of this approach, however, is enabled by selecting the “Batch” button on the main panel. Choosing this option once the pulmonary arterial curve has been designated puts the program into an automated processing mode, needing no additional user interaction. This mode works by scanning a box, of the chosen sample size, over successive, non-overlapping regions of an entire slice and calculating physiologic data for each sample, then repeating this for all slices of the data set. Samples not meeting user defined criteria (middle panel, Fig. 5) are rejected. Selection criteria under user control include the minimum and maximum arrival and mean transit times, minimum and maximum air, blood, and tissue fractional contents, and maximum chi squared. Criteria can easily be changed by selecting a parameter from the scrolling list and simply typing in the new value. Choosing a set of default criteria unique to the lung field limits user input to identifying the pulmonary artery, for input
to the blood flow model, and selecting the “Batch” button. Empirically derived selection criteria we normally use include air contents of 20-90% to eliminate airway structures, mean transit times between 4 and 10 s and arrival times between 1 and 10 s to eliminate major veins and arteries, blood contents of 5-40% and tissue contents of 20-60% to eliminate the chest wall and myocardium and heart chambers. The batch mode samples the image set for data collection only. The criteria are applied and the model is solved only when the user requests the computation of a functional image. This approach allows the program to respond quickly to changes in the region selection criteria without having to gather the data all over again. Another feature of the program is that it allows the functional image to represent ANY parameter of the model, not just flow. So it is possible to construct mean-transit-time images, peak-opacification images, regional tissue, blood, or air content images, etc. Blood flow calculations are made from a model of the user’s choosing (upper right, Fig. 5). Our program
Pulmonary structure and perfusion lE. A. HOFFMAN, J. K. TAJIK and S. D. KUGELMASS
Fig. 6. Color coded pulmonary blood flow (ml/gram parenchymafmin) image combining the UHRCT and HTRCT scans. Top row: Transverse images looking along the feet to head axis with the spine down. Middle row: Sagittal image sequence shown with the spine to the left in each cross section, and moving from right to left in the subject. Bottom row: Coronal image sequence moving from dorsal to ventral regions of the lung, with the right side of the animal shown on the viewer’s left in each coronal slice. Images were acquired from a dog in the supine position, and demonstrate a significant ventral to dorsal blood flow gradient (low blood flow is displayed in the blue end of the spectrum while the pink/white end of the scale represents regions of highest tissue flow). Higher perfusion is seen in the gravity dependent regions of the supine lung.
Fig. 7. UHRCT and HTRCT scans used in evaluating simulated pulmonary embolus. UHRCT images were used to visually locate the position of a balloon tip catheter in the lung field and then identify the vessel segment distal to the inflated balloon (lower right). Extracting an oblique section perpendicular to the vessel cross section, yields
109
110
Computerized
Medical
Imaging
and Graphics
currently provides myocardial blood flow models of Ritman (26) and Weiss (27) in addition to the Wolfkiel (23) model (shown selected in the panel), which we use for pulmonary perfusion calculations. Again, any blood flow model requiring a single input function and producing one output can be easily integrated into our program and added to a list of models for the user to choose. Bad time points can be omitted from the blood flow calculations by simply “clicking” on one of the time points listed in the “Use Phases” subheading on this panel. A text file containing the three dimensional coordinates of all accepted samples along with their calculated corresponding physiologic parameters is generated and can be saved for later analysis by any commercially available spreadsheet program. Color coded images of all physiologic parameters are also generated and may be viewed in a separate window (bottom right, Fig. 5). These images may also be saved to disk. A special feature of our module combines the complementary information from UHRCT and dynamic CT by automatically mapping color coded functional data into a high resolution volumetric scan of the same subject. If the subject was not moved between the two scan modes, and we know the exact scanner geometry and patient table location during the two scans, the functional and high resolution images can be easily combined. User interaction is limited to loading both data sets into memory and selecting the “High Res Image” button followed by selecting the “Apply” button (bottom right, Fig. 5). Any of the color coded functional data sets may be imbedded into the high resolution volume, allowing for highly detailed visualization of structure-function relationships. To display the combined structure-function image on an S-bit monitor (capable of reproducing 256 colors or shades of gray), during the merge process functional data are mapped onto the 0- I27 range while structural data are mapped onto the 128-255 range. Choosing the “split” color scale VIDA provides (O-127 range represented by color spectrum, 128-255 range by a gray scale), functional data appear in colors of the
January_February/1995,
Volume
rainbow while anatomic of gray on the monitor. EXAMPLES
19, Number
information
AND
1
appears in shades
APPLICATIONS
We have used the scanning protocol and analysis described above to characterize and visualize pulmonary blood flow distribution under various conditions, including alterations in body posture, lung volume, and with the imposition of artificially created pulmonary emboli. Fig. 6 depicts transverse, sagittal, and coronal slices of imbedded blood flow expressed as ml/gram parenchyma/min. Transverse images are shown looking along the feet to head axis with the spine down. The sagittal image sequence is shown with the spine to the left in each cross section, and moving from right to left in the subject. The coronal image sequence represents moving from dorsal to ventral regions of the lung, with the right side of the animal shown on the viewer’s left in each coronal slice. Images were acquired from a dog in the supine position, and demonstrate a significant ventral to dorsal blood flow gradient (low blood flow is displayed in the blue end of the spectrum while the pink/white end of the scale delineates regions of highest tissue flow). Higher perfusion is seen in the gravity dependent regions of the supine lung. However, when the animal is flipped to the prone posture this gradient does not reverse, but disappears, demonstrating important non-gravitational determinants of perfusion (28). In a separate experiment, depicted in Fig. 7, we simulated a pulmonary embolus by inflating a balloon catheter in the pulmonary vasculature. High spatial resolution tine CT (UHRCT) images of the lungs were used to visually locate the position of the balloon in the lung field and identify the vessel segment distal to the inflated balloon. After locating the vessel feeding the region of flow decrement (Fig. 7, lower right panel), we extracted an oblique section perpendicular to the vessel cross section, yielding the image shown in the lower left panel. Here one can see the balloon occlusion tip of the Swan-Ganz catheter (arrow). The balloon is also shown in cross section in the lower middle panel as identified via the plane’s intersection with the oblique section in the lower left panel. A high
the image shown in the lower left. Here one can see the balloon occlusion tip of the Swan-Ganz catheter (arrow). The balloon is also shown in cross section in the lower middle panel as identified via the plane’s intersection with the oblique section in the lower left panel. A HTRCT slice containing the distal vessel segment was then sampled (upper middle). Notice that the sample encompassing the region supplied by the distal vessel segment (2) has a flat time-intensity profile as compared with region 1 (sampled at the same chest level in the contralateral lung), indicating blockage of flow in region 2. The color coded image of regional blood flow (upper right) clearly shows this abnormal loss of flow, allowing easy embolus localization by visual inspection of the color coded images. Notice also the high detail visualization capabilities of UHRCT as compared to the blurry HTRCT image.
Pulmonary
structure
and perfusion
l
E. A HOFFMAN, J. K. TAIIK and S. D. KUGELMASS
temporal resolution tine CT (HTRCT) slice containing the distal vessel segment was then sampled. Parenchyma1 time-intensity curves of the sampled regions are shown in the figure. Notice that the sample encompassing the region supplied by the distal vessel segment (“2” in Fig. 7) has a flat time-intensity profile as compared with region 1 (sampled at the same chest level :~nthe contralateral lung), indicating blockage of flow in region 2. The color coded image of regional blood flow clearly shows an abnormal loss of flow, allowmg for easy embolus localization by visual inspection of the color coded images. Notice also the high detail visualization capabilities of UHRCT as compared to the blurry HTRCT image, due to partial volume effects (29) from the thick slice scanning mode The speed with which the physiologic data and corresponding color coded images are generated may prove useful for diagnostic radiology. A dynamic CT scan of a patient, for instance, could be analyzed interactively or using the automated mode of our software, providing relevant functional data with ease. Moreover, visual inspection of the corresponding color coded images provides a simple tool for easily detecting abnormal parameters (i.e., abnormally low or absence of blood flow in a region, uncharacteristically high local tissue or blood content, etc.) while the subject rests on the scanner bed. If a defect is suspected, a high spatial resolution scan can be focused on a particular region, to visualize the underlying anatomy, while limiting patient radiation exposure. Cine X-ray CT holds a unique position, to date, in being able to supply such a combination of structural and functional information of the lung from a single energy source. SUMMARY In this paper we have described a method for using two image acquisition modes, unique to tine X-ray CT scanners, for acquiring images of pulmonary anatomy and function. By using an appropriate blood flow model. regional parenchymal time-intensity curves, sampled from dynamic CT scans of bolus contrast injection, can be used to calculate regional blood flow, regional air, blood and tissue percentages, in addition to regional contrast mean transit times and arrival times. Functional data are calculated by a customized software module we have developed to automate and speed image analysis. User interaction with the software can be limited to designating the feeding vessel (i.e., pulmonary artery) for blood flow calculations. Color coded images representing any of the calculated functional data are generated at user request and can be automatically imbedded into a corresponding high
111
resolution scan of pulmonary anatomy. Our approach to image acquisition and post-processing can be used for detailed study of pulmonary structure-function relationships under various conditions in both clinical and research environments.
Acknowledgments-This paper appeared in part in Tajik et al., SPIE Proc. Conference 1905. 1993. This study was supported in part by NIH RO 1-HL42672.
REFERENCES 1. Tomioka, S.; Kubo, S.; Guy, H.J.B.; Prisk, G.K. Gravitational independence of single-breath washout tests in recumbent dogs. J. Appl. Physiol. 64:642-648; 1988. 2. Evans, J.W.; Wagner, P.D. Limits on VA/Q distributions from analysis of experimental inert gas elimination. J. Appl. Physiol. 42:889-898; 1977. 3. Milic-Emili, J.; Henderson, J.A.M.; Dolovich, M.B.; Trop, D.; Kaneko, K. Regional distribution of inspired gas in the lung. J. Appl. Physiol. 21:749-759; 1966. 4. Hughes, J.M.B.; Glazier, J.B.; Maloney, J.E.; West, J.B. Effect of lung volume on the distribution of pulmonary blood flow in man. Respir. Physiol. 4:58-72; 1968. 5. Reed, J.R. Jr.; Wood, E.H. Effect of body position on vertical distribution of pulmonary blood flow. J. Appl. Physiol. 28:303311; 1980. 6. Hoffman, EA.; Gefter, W.B.; Venegas, J. Frontier Pulmonary Imaging, In: Fishman, A.P., ed. Update: Pulmonary diseases and disorders. New York, NY: McGraw-Hill, Inc.; 1992: 323340. 7. Ritman, E.L.; Robb, R.A.; Harris, L.D. Imaging physiological functions: experience with the DSR. Philadelphia, PA: Praeger; 1985. 8. Hoffman, E.A. A historical perspective of heart and lung 3D imaging. In: Udupa, J.K.; Herman, G.T., eds. 3D Imaging in medicine. Boca Raton, FL: CRC Press; 1990: Chapter 11. 9. Liu, Y.H.; Hoffman, E.A.; Ritman, E.L. Measurement of threedimensional anatomy and function of pulmonary arteries with high-speed x-ray computed tomography. Invest. Radiol. 22:2836: 1987. 10. Liu, Y.H.; Ritman, E.L. Branching pattern of pulmonary arterial tree in anesthetized dogs. J. Biomed. Eng. 108:289-293; 1986. 11. Hoffman, E.A.; Sinak, L.J.; Robb, R.A.; Ritman, E.L. Noninvasive quantitative imaging of shape and volume of lungs. J. Appl. Physiol. 54:1414-1421; 1983. 12. Hoffman, E.A. Effect of body orientation on regional lung expansion: A computed tomographic approach. J. Appl. Physiol. 59:468-480; 1985. 13. Muller, N.L. Clinical value of high resolution CT in chronic diffuse lung disease. A. J. R. 157:1163-l 170; 1991. 14. D’Agincourt, L. Spiral scan capability increases utility of CT. Diagn. Imaging 13(3):98-104;.1991.. 15. Nauel. S.A.: Berain. C.J.: Paranioe, D.V.: Rubin. G.D. Maximum and minimim intensity p;djection of spiral CT data for simultaneous 3D imaging of the pulmonary vasculature and airways. Radiology 185(p): 126; 1992. 16. Glenny, R.W.; Robertson, H.T. Fractal properties of pulmonary blood flow: Characterization of spatial heterogeneity. J. Appl. Physiol. 69532-545; 1990. 17. Murata, K.; Itoh, H.; Senda, M.; Yonekura, Y.; Nishimura, K.; Irumi, T.; Oshima, S.; Torizuka, K. Stratified impairment of pulmonary ventilation in diffuse panbronchiolitis: PET and CT studies. J. Comput. Assist. Tomoar. 13:48-53; 1989. 18. Levine, D.N.; Hu, X.; Tan, K.K.;Galhotra, S.; Herrmann, A.; Pelizzari, C.A.; Chen, G.T.; Peck, R.N.; Chen, C.T.; Cooper, M.D. Integrated 3D display of MR, CT, and PET images of
112
19.
20. 21.
22.
23.
24.
25.
26.
27.
28.
29.
Computerized
Medical
Imaging
and Graphics
the brain. In: Udupa, J.K.; Herman, G.T., eds. 3D Imaging in medicine. Boca Raton, FL: CRC Press; 1990:Chapter 10. Gefter, W.B.; Kundel, H.; Hatabu, H. Magnetic resonance imaging in chest medicine. In: Fishman, A.P., ed. Update: Pulmonary diseases and disorders. New York, NY: McG;aw-Hill, Inc.; 1992:299-321. Boyd, D.P.; Lipton, M.J. Cardiac computed tomography. Proceedings of the IEEE. 71:298-307; 1983. Wei, H.; Hoffman, E.A.; Ritman, E.L.; Wood, E.H. Cardiogenic motion of right lung parenchyma in anesthetized intact dogs. J. Appl. Physiol. 58:384-391; 1985. Kalender, W.A.; Rienmueller, R.; Fichte, H.; Behr, J.; Beinert, T.; Seissler, W. Spirometrically gated CT measurement of lung density and structure. Radiology 185(p):354; 1992. Wolfiiel, C.J.; Ferguson, J.L.; Chomka, E.V.; Law, W.R.; Labin, I.N.; Tenzer, M.L.; Booker, M.; Brundage, B.H. Measurement of myocardial blood flow by ultrafast computed tomography. Circulation 76: 1262- 1273; 1987. Hoffman, E.A.; Gnanaprakasam, D.; Gupta, K.B.; Hofford, J.D.; Kugelmass, S.D.; Kulawiec, R.S. VIDA: An environment for multidimensional image display and analysis. SPIE Proc. 1660:694-711; 1992. Wood, S.; Hoford, J.; Hoffman, E.; Zerhouni, E.; Mitzner, W. Quantitative 3D reconstruction of airway and pulmonary vascular trees using HRCT. SPIE Proc. 1905: 316-323; 1993. Wong, T.; Wu, L.; Chung, N.; Ritman, E.L. Myocardial blood flow estimated by synchronous, multislice, high speed tomography. IEEE Transactions on Med. I. 8:70-77; 1989. Weiss, R.M.; Hajduczok, Z.D.; Marcus, M.L. A new tine CT algorithm for quantitation of myocardial perfusion. Circulation 78:11398; 1988. Larsen, R.L.; Bridges, C.R.; Beck, K.C.; Hoffman, E.A. Regional pulmonary blood flow via tine x-ray computed tomography. FASEB J. 4:A1074; 1991. Williams, G.; Bydder, G.M.; Kreel, J. The validity and use of computed tomography attenuation values. Br. Med. Bull. 36:279-287; 1980.
About the Author-Em A. HOFFMAN, Ph.D. is a cardiopulmonary physiologist who has used imaging based methods throughout his career in his studies of the heart and lung. He received his Ph.D. at the Mayo Graduate School of Medicine in conjunction with the University of Minnesota. He remained at the Mayo Clinic after graduation to take a post-doctoral position and later an Assistant
January-February11995.
Volume
19, Number
1
Professor appointment. While at Mayo, Dr. Hoffman was part of a research team which developed a one of a kind CT scanner known as the Dynamic Spatial Reconstructor used to acquire up to 240 contiguous sections of the body every 1160 second. In 1987, Dr. Hoffman moved to the Department of Radiology at the University of Pennsylvania as Assistant and later Associate Professor of Radiologic Sciences and Physiology. At Penn he headed the Section of Cardiothoracic Imaging Research within the Department of Radiology and formed a broad collaborative group of investigators (the Cardiothoracic Imaging Research Center or CIRC) representing cardiologists, biomedical engineers, pulmonologists, radiologists, pediatric cardiologists and radiologists, computer scientists, surgeons, etc. all joined with a common interest in using imaging and image processing to study physiologically important structure to function relationships. Under Dr. Hoffman’s direction, programmers within CIRC developed a broad based UNIX/C/X-windows based tool box for biomedical image processing known as VIDA for Volumetric Image Display and Analysis. In 1992, Dr. Hoffman and several members of his group at Penn moved to the University of Iowa wh’ere Dr. Hoffman now heads a Division of Physiologic Imaging where work on VIDA continues. At the University of Iowa College of Medicine. Dr. Hoffman is an Associate Professor of Radiology, Physiology, and Biomedical Engineering. Research interests include cardiac mechanics, pulmonary mechanics, and ventilation/perfusion relationships within the lung, and in general, the use of dynamic volumetric imaging to study physiology. More recently the lab has begun work to develop methods whereby rural hospitals can gather accurate volumetric image data sets of the thorax and abdomen on slower CT scanners and send the image data over the fiber network running throughout the state of Iowa for image analysis and interactive consultation. Dr. Hoffman has served as the organizer of SPIE’s Medical Imaging ‘94 Physiology and Function from Multidimensional Images Conference and is serving in this capacity again for Medical Imaging ‘95. About the AnthOr-JEHANGIR K. TAJIK received his Bachelor’s degree from St. Olaf College and is currently pursuing his PhD degree at the University of Iowa. His research interests span ventilation/perfusion matching, cardiovascular and pulmonary mechanics and quantitative medical imaging. About the Author-STEVEN KUGELMASS received his Ph.D. in Computer Science from Princeton University in 1988. From 1990 until 1992 he was Director of Computer Science at the Cardiothoratic Imaging Research Center at the Hospital of the University of Pennsylvania. In 1992 he became an assistant professor at the New Jersey Institute of Technology. He now works in private industry.