Radiological Physics and Technology

, Volume 1, Issue 2, pp 137–143

Development of functional chest imaging with a dynamic flat-panel detector (FPD)

Authors

    • Department of Radiological Technology, Graduate School of Medical ScienceKanazawa University
  • Shigeru Sanada
    • Department of Radiological Technology, Graduate School of Medical ScienceKanazawa University
  • Masaki Fujimura
    • Department of Cellular Transplantation Biology, Graduate School of Medical ScienceKanazawa University
  • Masahide Yasui
    • Department of Cellular Transplantation Biology, Graduate School of Medical ScienceKanazawa University
  • Kazuya Nakayama
    • Department of Radiological Technology, Graduate School of Medical ScienceKanazawa University
  • Takeshi Matsui
    • Department of RadiologyKanazawa University Hospital
  • Norio Hayashi
    • Department of RadiologyKanazawa University Hospital
  • Osamu Matsui
    • Department of Radiology, Graduate School of Medical ScienceKanazawa University
Article

DOI: 10.1007/s12194-008-0020-7

Cite this article as:
Tanaka, R., Sanada, S., Fujimura, M. et al. Radiol Phys Technol (2008) 1: 137. doi:10.1007/s12194-008-0020-7

Abstract

Dynamic FPD permits the acquisition of distortion-free radiographs with a large field of view and high image quality. In the present study, we investigated the feasibility of functional imaging for evaluating the pulmonary sequential blood distribution with an FPD, based on changes in pixel values during cardiac pumping. Dynamic chest radiographs of seven normal subjects were obtained in the expiratory phase by use of an FPD system. We measured the average pixel value in each region of interest that was located manually in the heart and lung areas. Subsequently, inter-frame differences and differences from a minimum-intensity projection image, which was created from one cardiac cycle, were calculated. These difference values were then superimposed on dynamic chest radiographs in the form of a color display, and sequential blood distribution images and a blood distribution map were created. The results were compared to typical data on normal cardiac physiology. The clinical effectiveness of our method was evaluated in a patient who had abnormal pulmonary blood flow. In normal cases, there was a strong correlation between the cardiac cycle and changes in pixel value. Sequential blood distribution images showed a normal pattern at determined by the physiology of pulmonary blood flow, with a symmetric distribution and no blood flow defects throughout the entire lung region. These findings indicated that pulmonary blood flow was reflected on dynamic chest radiographs. In an abnormal case, a defect in blood flow was shown as defective in color in a blood distribution map. The present method has the potential for evaluation of local blood flow as an optional application in general chest radiography.

Keywords

Functional imagingDynamic chest radiographyFlat-panel detectorBlood flowVisualizationImage subtraction

Introduction

Pulmonary blood flow mirrors pulmonary and cardiac physiology [14] by showing: redistribution or cephalization of pulmonary blood flow indicating the presence of pulmonary venous hypertension [5], a centralized pulmonary blood flow pattern indicating pulmonary arterial hypertension [6], or widening of the vascular pedicle indicating an increase in the circulating blood volume [7]. Therefore, when interpreting a chest radiograph from a cardiac perspective, it is important for a radiologist first to determine the type of pulmonary blood flow pattern that is present on the chest radiograph. Circulation dynamics are also reflected on fluoroscopic images as changes in X-ray translucency. There have been many reports showing the feasibility of pulmonary densitometry [811]. However, these methods have not been adopted for clinical use because of technical limitations, such as poor image quality and a small field of view (FOV) [12]. Moreover, it is extremely difficult to evaluate slight changes in X-ray translucency by visual observation.

At present, pulmonary blood flow is evaluated by lung perfusion scintigraphy, perfusion computed tomography (CT) [1316], and magnetic resonance imaging (MRI) [1721]. Although these examinations are useful for assessing the pulmonary blood flow pattern, they are not simple procedures. If functional information becomes more readily available, it will be very helpful for determining an appropriate examination procedure and for patient follow-up.

Recently developed dynamic flat-panel detectors (FPDs) can provide distortion-free radiographs with a large field of view and high image quality compared with radiographs obtained with an image intensifier (I.I.) system. Dynamic FPDs are expected to be used more widely in the near future. Moreover, dynamic chest radiography combined with computer analysis would allow a quantitative assessment of circulation dynamics. In previous work, we developed functional chest radiography by using a dynamic FPD and computerized methods of evaluating respiratory kinetics, such as diaphragmatic movement and regional ventilation [2224]. In a clinical study, areas with ventilation abnormality were indicated as decreased areas in changes in pixel value during respiration [25]. In this study, we focused on pulmonary and cardiac blood flow. We investigated the feasibility of functional imaging with a dynamic FPD, and we developed a method for analyzing and visualizing the results as relative pulmonary blood flow. Here, the preliminary results of a clinical study are described in terms of the physiology of blood flow dynamics.

Materials and methods

Image acquisition

Dynamic chest radiographs of seven normal subjects were obtained in the expiratory phase with a modified FPD system (CXDI-40C; Canon Inc., Tokyo, Japan), in combination with electrocardiograms (ECGs) recorded with a Holter monitor (Digital walk, FM-120; Fukuda Denshi Ltd, Tokyo, Japan). An X-ray pulse signal was obtained from the FPD system by use of an oscilloscope (two-channel color digital oscilloscope, TDS3012B; Tektronix Ltd, Tokyo, Japan). The modified FPD was an indirect type made of CsI, and was capable of taking images at up to 6 frames/s, which is the highest rate possible with our system at present. Exposure conditions were as follows: 110 kV, 80 mA, 6.3 ms, source-image distance 2 m, 6 frames/s, 2-mm Al filter. Twenty-four posteroanterior (PA) chest images were obtained in 4 s. The entrance surface dose for 24 frames, measured in air without backscatter, was approximately 0.32 mGy, which was 1.3-fold greater than that of conventional PA chest radiography by use of a Fuji Computed Radiography system (Fuji Medical Systems Co, Ltd, Tokyo, Japan) in our hospital. The matrix size was 672 × 672 pixels, the pixel size was 640 × 640 μm, and the gray-scale range of the images was 4096. Low pixel values were related to dark areas in the images, and these, in turn, were related to high X-ray translucency in this system. The pixel value was inversely proportional to the logarithm of the incident exposure in the FPD. Approval for the study was obtained from our institutional review board, and the subjects gave written informed consent prior to participation.

Image analysis

Image analysis was performed on a personal computer (operating system, Windows 2000, Microsoft, Redmond, WA, USA; CPU, Pentium 4, 2.6 GHz; Memory, 1 GB) with our algorithm described below (development environment: C++Builder; Borland, Scotts Valley, CA, USA), for determining the cardiac phase in each frame and measuring heart wall motion and changes in pixel value on dynamic chest radiographs.

Determination of the cardiac phase in each frame

The cardiac phase in each frame was determined based on two signals, the ECG and the X-ray pulse wave, obtained by oscilloscope. We analyzed the ECG to determine the R wave and the T wave, because these are important indices for understanding the cardiac phase. The cardiac phase consists of two main phases: the ventricular systole phase from the peak of the R wave to the end of the T wave and the remaining time, which is considered the diastole phase, as shown in Fig. 1. The X-ray pulse wave was also analyzed for determination of the timing of the imaging.
https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig1_HTML.gif
Fig. 1

Electrocardiogram (ECG) and X-ray pulse wave. One cardiac phase consisted of the ventricular systole phase from the peak of the R wave to the end of the T wave; the remaining time is the diastole phase

Measurement of changes in pixel values and cardiac motion

We measured the average pixel value in each region of interest (ROI), which was located manually on the ventricles, atria, aortic arch, pulmonary arteries, pulmonary vein, peripheral-lung blood vessels, and reference areas, as shown in Fig. 2, to investigate the difference in each area with known changes in blood flow. The reference area for measuring image noise without the influence of heartbeat and respiration was located in the shoulder and in air. The ROIs were squares with 7 pixels on a side, and were nearly equivalent to 1 square cm, which was almost the same size as the intercostal area on the images. The results were compared to typical data from normal cardiac physiology, such as blood pressure and blood volume, with ECG data used as reference [26].
https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig2_HTML.jpg
Fig. 2

Measurement location. Small squares show ROIs for measuring average pixel value, and the horizontal line shows a profile for measuring left-ventricle motion (a left ventricle (LV); b right ventricle (RV); c aortic arch (AA); d left atrium (LA); e right atrium (RA); f left pulmonary artery (LPA); g right pulmonary artery (RPA); h left pulmonary veins (LPV); i right pulmonary veins (RPV); other ROIs, peripheral vessel; ref. 1 shoulder; ref. 2 air)

We also calculated the rate of change in the average pixel value to investigate differences between the locations. We hypothesized that each area has the same rate of change and tested this hypothesis by one-way analysis of variance and Tukey test [27].

In addition, we measured cardiac motion by analyzing a profile automatically located on the left ventricle, as shown in Fig. 2. The results were also used for evaluating the average pixel value in each ROI.

Visualization of changes in pixel values

The lung area was determined by edge detection with use of the first-derivative technique and an iterative contour-smoothing algorithm, as described in detail elsewhere [28, 29]. Subsequently, the inter-frame difference was determined throughout all frames. Sequential blood distribution images were then created by superimposing of difference values on dynamic chest radiographs in the form of a color display, by use of a color table in which positive changes (lower X-ray translucency) were shown in warm colors and negative changes (higher X-ray translucency) were shown in cool colors. In addition, a minimum-intensity projection (MINIP) image was created in one cardiac cycle, which was composed of pixel values showing the least blood during one cardiac cycle. Temporal subtraction was performed between a MINIP image and a frame at the end of the systole phase. The blood distribution in one cardiac cycle was also created by superimposing of the difference values on dynamic chest radiographs in the form of a color display. The results were compared to the typical normal circulation [26].

Assessment of clinical usefulness

One patient with pulmonary fibrosis (56-year-old female) was examined by use of the same protocol for evaluating the clinical usefulness of our method. Our results were compared with the findings in chest radiographs, CT, and perfusion scintigraphy of the lung.

Results

Figure 3 shows changes in pixel value measured in a normal subject. There was a strong correlation between the cardiac cycle and changes in pixel value, which were measured in the ventricles, atria, aortic arch, and pulmonary arteries. The other areas also changed in synchronization with cardiac pumping. In all normal subjects, pixel values measured in each ROI showed the following tendencies: In the ventricular systole phase, pixel values in the ventricles decreased. In contrast, those in the aortic arch and pulmonary arteries rapidly increased and then decreased gradually until the next systole phase. In the ventricular diastole phase, pixel values in the atria decreased rapidly, whereas those in the ventricles increased. Thereafter, pixel values in the atria increased gradually until the next diastole phase. Pixel values in the lung area increased slightly in the systole phase and decreased in the diastole phase.
https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig3_HTML.gif
Fig. 3

Relationship between cardiac cycle and changes in pixel values. a Normal circulation determined by cardiac physiology, b measured pixel values in each ROI in the present study (normal, 22-year-old man)

Figure 4 shows the average rate of change in pixel value for all subjects. The results measured in each ROI decreased in the following order: left ventricle > left atrium > aortic arch> right atrium > right ventricle > left pulmonary artery. The rate of change in pixel value was 1.14% in the left ventricle, those in the other heart areas ranged from 0.62 to 0.83%, those in other vessels ranged from 0.34 to 0.67%, those in the lung areas ranged from 0.18 to 0.38%, and those in the reference area without the influence of cardiac pumping were 0.11%, as shown in Fig. 4. There was no significant difference between some lung areas and the reference area.
https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig4_HTML.gif
Fig. 4

Average changes in pixel values measured in each ROI. Error bars show ±SD (n = 7). (SD standard deviation, NS not significant)

We also succeeded in visualizing slight changes in pixel value as sequential blood distribution images and as blood distribution maps without contrast media (Fig. 5). In all normal subjects, the sequential blood distribution images showed a normal pattern determined by the physiology of pulmonary blood flow, which diffuses from around the pulmonary arteries to the peripheral area. The blood distribution map also indicated a normal pattern of circulation, with no perfusion defects throughout the entire lung region. Figure 6 shows results in a patient with lung fibrosis (56-year-old female). The chest radiograph and CT (coronal section) showed advanced fibrosis of the lung in the upper area (Fig. 6a, b). The results of perfusion scintigraphy of the lung showed some areas of decreased blood flow, as seen in Fig. 6c. The blood distribution map showed some areas with decreased changes in pixel value (Fig. 6d), which were coincident with the abnormal area in perfusion scintigraphy of the lung.
https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig5_HTML.gif
Fig. 5

Computer scheme for creating sequential blood distribution images and blood distribution map

https://static-content.springer.com/image/art%3A10.1007%2Fs12194-008-0020-7/MediaObjects/12194_2008_20_Fig6_HTML.jpg
Fig. 6

Results of our method and the other clinical examinations in a subject with pulmonary fibrosis (56-year-old female). a Chest radiograph, b computed tomography (coronal section), c lung perfusion scintigraphy, and d blood distribution map. In the blood distribution map, there were some reductions of changes in pixel value (solid-line circles), which were consistent with defect in blood flow as indicated by scintigraphic test (broken-line circles in c)

Discussion

The changes in pixel values measured in each ROI can be explained by normal circulation dynamics as indicated below: (i) At the end of the diastole phase, the ventricles are at the maximum volume, as shown by large pixel values in the ventricles. (ii) In the early ventricular systole phase, from closure of the atrioventricular (AV) valves to opening of the aortic valve, the ventricular volume remained constant, shown as the absence of a significant change in pixel values during this period. (iii) After opening of the aortic arch, blood is pumped from the ventricles into the aortic arch and pulmonary arteries. This was shown as a decrease in pixel values in the ventricles and an increase in pixel values in the aortic arch and pulmonary arteries. (iv) In the late ventricular systole phase, ventricular volume and aortic blood flow decrease, shown as a continuous decrease in pixel values. (v) In the early ventricular diastole phase, from closure of the aortic valve to opening of the AV valves, the ventricular volume remains constant. Thus, there was no significant change in the pixel value. (vi) Blood rapidly moves from the atria to the ventricles in response to opening of the AV valves; this is why the pixel values in the ventricles increased while, in contrast, those in the atria and pulmonary veins decreased.

Although there was no significant change in the measured pixel values compared to those in the heart and large vessels, the pixel values measured in the lung areas, i.e., the peripheral pulmonary vessels, changed in synchronization with cardiac pumping. These findings indicated that pulmonary blood flow was reflected on dynamic chest radiographs, and that the present method has the potential to evaluate local blood flow. Further investigations are required for relating the findings to clinical parameters, such as cardiac output and ejection fraction. In addition, the procedure took a few minutes for the analysis of one subject. Although the aim of this study, to investigate the relationship between the changes in pixel value and cardiac physiology, was achieved, there would be intra- and inter-observer variability in the results. Therefore, it is one of the future tasks to develop a computerized method for the definition of reference ROIs.

Sequential blood distribution images and blood distribution maps were very useful for interpreting slight changes in pixel values. Furthermore, the large FOV allows us to evaluate the entire lung area. Sequential blood distribution images were created by inter-frame differences in pixel values due to cardiac pumping. Thus, these images can provide information on blood flow velocities during each cardiac phase. In addition, blood distribution maps were created by differences in pixel values between a frame in the minimum blood volume during one cardiac cycle and a frame at the end of the systole phase. Thus, the image can provide information on relative blood volumes during one cardiac cycle. The results in an abnormal case showed that abnormalities, such as a decrease or defect in blood flow, would be shown as defects in color.

These findings indicated that dynamic chest radiography with a dynamic FPD has potential for functional imaging. The present method is expected to be a rapid and simple method for evaluation of blood flow in general chest radiography, because FPDs will soon be more widely available. However, there is not enough clinical evidence supporting the usefulness of the present method. Further studies in larger numbers of subjects with abnormalities in blood flow, such as pulmonary embolism, pulmonary hypertension, pulmonary edema, and general heart diseases, are required along with investigations into the ability of this method to detect abnormalities. The errors due to vessel misalignment and the reproducibility of the present method should also be addressed. In addition, imaging was performed at 6 frames/s due to technical limitations in the present study. For the precise evaluation of blood flow, dynamic chest radiographs should be obtained at a higher rate, while maintaining an acceptable total exposure dose to the patient. Furthermore, there were concerns about the effect of body motion, respiration, dilation and contraction of the vessels themselves, as well as image noise. In the present study, there was no significant influence on the measurement of pixel values due to artifacts resulting from body motion and respiration. The change in pixel value resulting from cardiac pumping showed a cyclic pattern. Thus, these changes could be separated from noise and the other factors. However, it is necessary to develop a computer algorithm to reduce the artifacts in patients who cannot hold their breath. In addition, dilation and contraction of vessels would not be considered to affect the measurement of pixel values in projected images with a relatively large size because the rate of change is reported to be approximately ±10% [26].

Although the present method lacks 3D anatomic information, it is expected to serve as a rapid and simple method for evaluating blood flow. In addition, estimation of the cardiac phase based on heart wall motion will further simplify the present method of ECG monitoring. This method can be used as a rapid substitute for lung perfusion scintigraphy, perfusion CT, and MRI, and could be applied to cone beam CT and 4D CT, as an optional means of evaluating blood flow.

Conclusions

We investigated the feasibility of functional imaging with an FPD, based on changes in pixel values during cardiac pumping. Pulmonary blood flow was reflected on dynamic chest radiographs, and the inter-frame subtraction technique was very useful for interpreting slight changes in pixel values on dynamic chest radiographs. Dynamic chest radiography overcomes some of the limitations of I.I. methods and increases the diagnostic information in general chest radiography. FPD will soon be more widely available; thus, the present method is expected to be a rapid and simple one for evaluating pulmonary blood flow without contrast media in general chest radiography.

Acknowledgments

The authors are grateful to the volunteers and to Yasuhiro Yamauchi at Fukuda Denshi Co., and the technologists of the Dept. of Radiology, Kanazawa University Hospital, who assisted with data acquisition. We thank Kunio Doi, Ph.D. and researchers at the University of Chicago for valuable discussions regarding image analysis. The present study won 1st prize as a poster presentation in Computer-assisted Radiology and Surgery (CARS) 2006. The authors thank the editors and reviewers who spent a great deal of time and gave us informative advice for improving our manuscript. This work was supported in part by the Nakajima Foundation, Konica Minolta Imaging Science Foundation, and a Grant-in-Aid for Scientific Research from the Japanese Ministry of Education, Culture, Sports, Science, and Technology.

Copyright information

© Japanese Society of Radiological Technology and Japan Society of Medical Physics 2008