Background
Chronic obstructive pulmonary disease (COPD) is characterised by obstructed airways and parenchymal destruction. Characteristically, varying degrees of abnormalities are found in different parts of the COPD lung, and some parts of the lung may even be normal. Consequently, abnormal ventilation distribution is the first abnormality to be detected in the early stages of the disease [
1].
COPD starts with inflammation and obstruction in peripheral airways [
2,
3] where the resistance is very low [
4]. Therefore, conventional lung function tests (spirometry) are insensitive [
5]. Imaging techniques primarily aim at localising lesions, but as these techniques become digital, new interpretative possibilities are now arising. By using high-resolution computed tomography (HRCT), mild emphysema may be assessed by quantification of the density distribution [
6]. Although computed tomography is an excellent method for identifying anatomical changes in the lung tissue, it provides little information about lung function reduction [
6]. In contrast, lung function reduction is both imaged and assessed directly by lung single-photon emission computed tomography (lung SPECT), a method often used in diagnosis of pulmonary embolism [
7,
8]. Hence, previous reports have described ventilation and perfusion SPECT as a sensitive method of detecting early changes in COPD [
9]. Moreover, SPECT findings correlate significantly with emphysema scored by HRCT and lung function tests [
9].
It is common to interpret lung SPECT images qualitatively. However, quantitative information obtained from a SPECT examination has the potential to provide much more information. Different quantitative methods for assessment of lung SPECT images have been reported [
10‐
16]. Xu et al. [
11] quantified inhomogeneities in ventilation SPECT images of COPD patients using 50 MBq of Technegas (Vita Medical Limited, Sydney, Australia), measuring the coefficient of variance (CV) in the lung elements. In that study, transaxial SPECT slices (1-cm thickness with 3.5-cm spacing) were acquired using a low-energy, general-purpose collimator, filtered back-projection with non-homogeneous attenuation correction and a 2D Hann post-filter. These were then divided into 2 × 2 × 1 cm
3 elements. By that method, it was possible to separate non-smoking healthy subjects and ‘healthy smokers’ from COPD patients. Importantly, however, that method was not sensitive enough to discriminate between healthy non-smokers and healthy smokers. Our overall aim is to develop a quantitative method sensitive enough to discriminate between these two groups. Therefore, the present study is currently followed by a larger clinical study evaluating the ability of the method used in this paper to differentiate between healthy non-smokers and healthy smokers.
The present study had two aims. Firstly, we wanted to develop a quantitative method using Monte Carlo simulated lung SPECT images of phantom lungs that could discriminate between uniform (healthy, simulated normal) and non-uniform (non-healthy, simulated COPD) activity distributions corresponding to COPD lung changes of varying severity. Secondly, we wished to evaluate the ability of the same method in a clinical environment to differentiate between human subjects with advanced COPD and healthy volunteers.
Discussion
In the present study, we have modified a method by Xu et al. in order to improve the quantification of ventilation inhomogeneities in a phantom model of a COPD lung. In contrast to what has been previously reported, our improved method was able to assess even minor COPD changes by using the AUC(CVT) value, as a global value of ventilation inhomogeneities, and to discriminate these changes from a model of a healthy homogeneous lung. The present pilot study also shows that our way of performing lung SPECT and calculating the AUC(CVT) significantly discriminates non-smoking healthy volunteers from patients with advanced COPD.
The NCAT software is able to create thorax voxel phantoms of a human, based on a fine-segmented male [
32]. The NCAT software is flexible since different sizes and shapes of different tissues can be selected and natural movements caused by heartbeat and respiration can be modelled. Because most of the COPD patients are elderly [
33], we decided to use a lung volume corresponding to a 65-year-old male. Low-ventilated regions, associated with anatomical changes of COPD, are distributed in patients in various ways, and these volumes can vary between 0.5 mm (the size of a few alveoli) and several centimetres in diameter. One of the aims of the present study was to mimic mild to moderate changes of COPD, and therefore small lesions were of interest. Since the spatial resolution of the SPECT system is about 1 to 1.5 cm (expressed in FWHM), the lesions modelled had a diameter of 1 and 2 cm. Lacking previous studies on the distribution of ventilation inhomogeneities in mild COPD, we assumed that COPD lesions are either evenly distributed in the whole lung volume or centred in clusters (see Table
1). In this way, our method was evaluated on two groups with completely different lesion distributions. The density of the lesions was approximated to be the same as for the healthy lung tissue since we aimed to model less severe changes of COPD. Indeed, activity distribution 1 cm
50% 10% (distribution 2 in Table
1) illustrates activity inhomogeneities that are almost too small for the SPECT system to resolve, and, clearly, not visible to the human eye in the reconstructed image (
cf. Figure
5). Thus, we believe this distribution would be a good representative of a case of mild COPD. However, it should be pointed out that the activity distributions selected in the different cases were not primarily chosen because of their consistency with biology, but more because of our ability to unambiguously describe them.
How does the magnitude of the present volume of ventilation defects compare to reductions of spirometric variables? Fifty percent reduction of the ventilation in 10% of the lung volume (distributions 2 and 6 in Table
1) corresponds to the 5% total reduction of the functioning lung tissue. Although not exactly comparable, reductions of spirometric variables of similar magnitudes are likely to remain undetected, as the normal range of spirometric variables is roughly ±15% to 20%. Thus, in our anthropomorphic phantom, we consider most of the present volume of ventilation defects as comparatively small.
When considering the result of the presented quantitative method, a number of important parameters have to be accounted for, e.g. count density (statistical noise), collimator, number of iterations and subsets, reconstruction compensations and post-filtering. In this work we used clinically relevant values for these parameters. Other important factors are the method-specific ones, i.e. the lung edge effects and kernel size that we evaluated. The resolution of the SPECT system is limited, and therefore, high CV values will always be found in the periphery of the healthy lung. Lung edges with low activity lesions will instead give lower CV values. High CV values due to healthy edges reduce the differences between frequency functions from healthy and unhealthy activity distributions; i.e. they also reduce the separation between corresponding AUC(CV
T) values. Therefore, the CV analysis was performed in a volume that had part of the edge effect excluded. Exclusion of a one-voxel layer, however, removes 21% of the phantom lung and 20% to 29% of the human lung parenchyma from the analysis, which is why small lesions in the periphery might not be detected. The kernel approach used does not exclude any additional volume of the lung in the analysis. Kernels with five different side lengths were evaluated in the simulated COPD study, i.e. 1.0, 1.7, 2.3, 3.0, 3.6 and 4.3 cm. Increasing the side length resulted in an increasing differentiation between the AUC(CV
T) values of the simulated normal and 1 cm
50% 10% evenly distribution (distributions 1 and 2 in Table
1). However, for the two largest cube sizes tested, in combination with some values of the above-mentioned parameters (e.g. LEHR collimator, a Butterworth post-filter with a cut-off frequency of 0.3 cm
−1 and a power of 6, 125 MBq, ten iterations and 16 subsets and no exclusion of edge voxels), the CV frequency functions of the simulated normal distribution were double peaked. Therefore, to minimise the risk of double peaks in a clinical setting, we chose a kernel with the side length of 3 cm, including 729 voxels.
The present method of quantitative analysis has two major advantages. Firstly, it discriminates cases with the same loss of ventilation, but with inhomogeneities differently distributed and with different lesion sizes, from the simulated normal lung, even when COPD changes are minor. For example, distributions 2 (1 cm
50% 10% evenly), 3 (1 cm
50% 12% centred) and 6 (2 cm
50% 10% evenly) in Figure
3 all represent only a 5% to 6% total reduction of ventilation, but their resulting AUC(CV
T) values are well above the value of the normal lung. The same holds true for distributions 4 (1 cm
0% 10% evenly), 5 (1 cm
0% 12% centred) and 7 (2 cm
0% 10% evenly), which all correspond to a 10% to 12% total reduction of ventilation. Secondly, for the same lesion size, increasing AUC(CV
T) values tend to correlate with decreasing total ventilation. For example, distribution 6 (2 cm
50% 10% evenly) in Figure
3, corresponding to a 5% total reduction of ventilation, gives a lower AUC(CV
T) value than distribution 7 (2 cm
0% 10% evenly) corresponding to a 10% total reduction of ventilation. But clearly, even with the same loss of ventilation, activity distributions with a few large lesions (≥1.5 cm) with low activity will give higher AUC(CV
T) values than activity distributions with many small lesions (≤1.5 cm) with relatively high activity in each lesion (e.g. see Figure
3 and Figure
5, distributions 4 and 7 at rows 3 and 5, column C).
It is desirable that distributions with the same total reduction of ventilation result in the same level of AUC(CV
T) values independent of lesion shape and distribution. This is the case of the evenly distribution 4 and the clustered distribution 5 (with a total reduction of ventilation of 10% to 12%). However, the difference in distribution can be seen in the frequency functions (Figure
4) and in corresponding CV matrices (Figure
5). Clusters positioned in other parts of the lung have not been investigated. Factors influencing the result are the volume and activity concentration of the spheres in the cluster and the size of the cluster’s surface area towards the uniform part of the lung, and not where in the lung the cluster is positioned.
One limitation of the present method is its loss of sensitivity when COPD changes become more advanced. An activity distribution with an AUC(CVT) value close to 100% can easily be separated from the simulated normal distribution; however, further reduction of ventilation will only result in an almost unchanged AUC(CVT) value (because 100% is the highest possible value). On the other hand, the present method is customised to detect early and minor COPD changes and not to be a diagnostic for advanced emphysema, for which methods such as HRCT are more useful. Furthermore, in cases of advanced COPD, the appearance of the frequency function of CV values can be used directly, without calculating the CVT and the AUC(CVT), to estimate the COPD severity.
The activity distribution in healthy humans is not as homogeneous as the simulated normal distribution used in the simulated COPD study, which might be due to the shape of the bronchial tree and the gravity influencing the lung. This difference is seen in the defined CV
T values, 20.5% for the simulated COPD study and 22.0% for the human study. Furthermore, the human thorax exhibits a large variety of sizes and shapes. In order to decrease the influence of the lung size, AUC(CV
T) values are presented in a percentage of total lung volume instead of absolute values. This prevents large healthy lungs giving high AUC(CV
T) values and small inhomogeneous lungs giving low AUC(CV
T) values. However, a larger 95% CI of the AUC(CV
T) value of the healthy volunteers of ±13% around the mean compared to the simulated COPD study of ±1% was found. This large variation of the healthy volunteers might be due to genetic variations, age effects, different histories of occupational and environmental exposures of noxious particles and gases e.g. passive exposure to tobacco smoke and varying techniques of inhaling the Technegas. The total amount of inhaled Technegas, expressed in megabecquerel, also affects the resulting CV values. A reconstructed ventilation distribution based on a low activity level will contain higher statistical noise compared to a distribution based on a higher activity level. Higher noise levels will result in higher CV values. A variation in the amount of inhaled Technegas was observed between the human subjects. Therefore, due to different activity levels, small shifts along the CV axis of the frequency functions in Figure
7 are present. Procedures resulting in more reproducible amounts of inhaled Technegas and normalisation methods that after image acquisition can compensate for such shifts will be further investigated. Reproducibility may be improved also by control of the depth and inhalation flow rate of the Technegas administration. In particular, the inhalation flow rate influences the ventilation distribution in normal subjects [
34] and almost certainly also the deposition in patients with COPD.
The activity distribution in the COPD patients was characterised by areas of low activity and areas of high activity i.e. hot spots. Low activity areas are caused by reduced regional ventilation resulting in low Technegas particle deposition. Reduced regional ventilation may be caused by local obstruction of peripheral airways and/or by emphysematous areas with low elasticity. Hot spots on the other hand, as pointed out by Pellegrino et al. [
35], may be caused by obstruction of central airways probably resulting in some Technegas impaction during inspiration but in particular by facilitating regional airflow limitation during expiration. Airflow limitation implies pronounced and oscillating narrowing of airway walls causing strong turbulence resulting in high impaction of Technegas particles during expiration [
35]. These hot spots were included in the assessment of the CV values. This might be considered as a weakness of the present method, since the intention behind the use of CV values was to identify low activity volumes as an indication of disease and not to generate high CV values due to abnormal uptake. On the other hand, such hot spots are typical of the COPD lung and might therefore be accepted as contributing to the analysis. How these hot spots influence the analysis will be evaluated in our future work. Presently, we investigate if the present method is reliable enough for detection of COPD changes in smokers without manifested COPD.
Competing interests
The authors declare that they have no conflict of interest.
Authors' contributions
PN participated in study planning, selected healthy controls, performed all lung SPECT analysis, evaluated data and wrote the manuscript. HLP participated in planning of the human study, selected patients, evaluated the data and wrote the manuscript. GAC and BB participated in planning and writing. MK participated in writing. MS and AG planned the study, evaluated data, oversaw the writing, and provided grant funding. All authors read and approved the final manuscript.