We would like to thank Drs. Hochhegger, Irion and Marchiori for their kind comments and their thoughtful questions. The issue regarding appropriate density-mask thresholds for the characterization of emphysema via quantitative computed tomography (CT)-based assessment of the lung is an important one. Quantitative CT has been used with increasing frequency to assess presence, extent, and location of emphysema-like lung for the purposes of pharmaceutical efficacy testing, safety testing, population selection for device placement, and for the identification of homogeneous subphenotypes within the chronic obstructive pulmonary disease (COPD) population to better target new therapeutic development efforts. Although quantitative CT has contributed greatly to these efforts, the measures are, in our opinion, not exact enough to worry about whether or not to set a threshold at -950, -960, or -970, to identify emphysema-like lung, particularly when such thresholds are selected through a comparison with excised formalin fixed lung tissue and a single scanner and reconstruction kernel. Extensive phantom studies are in the process of being carried out in conjunction with several National Institutes of Health–supported multicenter trials including the Severe Asthma Research Program (SARP) COPD-Gene, and SubPopulations and InteRmediate Outcome Measures in COPD Study (SPIROMICS). Data are being gathered into manuscript form, but the general findings are that Hounsfield units for air within a phantom or person vary among manufacturers and among models from the same manufacturer and between reconstruction kernels within and between manufacturers. Furthermore, it is well recognized that effort variations during an end inspiratory breath hold also change lung density values. Yet another issue is that as noise increases with decreasing x-ray dose (mAs), the noise at the low-density range of the Hounsfield scale is skewed by the fact that many CT manufacturers clip Hounsfield units at -1024 HU. This skewing of the noise alters mean lung density.
Having said all of this, if one looks at the cumulative density histogram (plotting percent voxels falling below a given HU on the y axis and HU on the x axis) for the lung field of a given patient group (COPD GOLD classification based on FEV1/FVC), there is a linear shift upwards with increasing severity of COPD as shown in Figure 1 . The implication of this is that if one compares populations, one will see a difference at almost any selected density mask threshold. Of importance is to be consistent and to assure that study protocols serve to minimize the various sources of error.
Open full size image
Figure 1
Cumulative density histograms formed out of the means of groups of subjects imaged via a lung cancer screening protocol at the University of Iowa. Subjects were grouped based on the results of their pulmonary function tests according to the GOLD classification scheme, whereby GOLD1 has the most mild disease and GOLD 4 has the most severe. This graph is modified from that presented in abstract form .
In the study referenced by the letter from Hochhegger and colleagues, we chose to use thresholds of -850, -910, and -950 and focused our discussion on -910 HU. The goal of our study was to understand the validity and reproducibility of density-mask measures obtained from scans performed to assess coronary calcium in the Multi-Ethnic Study on Atherosclerosis (MESA) Lung Study. These three values have been used widely in many studies of emphysema and derive from the work of Coxson et al in which pathology specimens were compared with CT scan values and Hounsfield units were selected that best represented lung specimens judged to fall within ranges of mild, moderate and severe emphysema. (We recognize that there was subsequently an erratum to Coxson’s work pointing out that some data in the study were gathered post contrast delivery; however, the authors discuss that the inclusion of these data did not adversely affect the conclusions of their work.) We rounded the -856 value of Coxson et al to -850 because we do not feel that the precision of the original work warrants a suggestion that one need to worry about the thresholds down to a single Hounsfield value. The value of -910 was selected as our primary measure in the MESA population because the presence of more severe forms of emphysema was rare. However, we evaluated data for both -910 and -950 thresholds and reported that the conclusions were similar for each. The paper by Madani et al , which is cited in the letter to the editor, performed a comparison of CT metrics with microscopic morphometry similar to that performed by Coxson et al . Neither CT nor microscopic-based morphometrics are free from controversy. There has recently been considerable debate regarding the best method for fixation of the lung when assessing alveolar morphometry and the best method for quantifying the resultant morphometry post fixation. Formalin fixation is known to cause considerable shrinkage and distortion of the lung. Although postmortem analysis of lungs in comparison with CT metrics is important for the establishment of general correlations between density measures and specific pathology, it is most likely not meaningful to use a single study, performed on a single scanner with a single CT slice thickness and reconstruction kernel tied to a microscope-based analysis that used formalin fixation. We understand that this is what is being done with the work of Coxson et al; however, the values survive both because of precedent as well as the fact that the values continue to be found to provide values that yield the most significant results in studies which have looked at multiple thresholds including our own. The work of Madani et al may be useful for determining a threshold that links CT density with a particular pathologic profile in the particular study performed at their site, but the notion that this one study establishes -960 or -970 to identify true emphysematous lung is likely not correct (any more than is Coxson’s -910 value an absolute standard). Even if CT measures were stable across sites, manufacturers and models, resistant to subject specific beam-hardening and scatter and all subjects held their breath during scanning at a well standardized lung volume, emphysema is a continuum with there being anywhere from a few alveoli or ascini effected to the extreme of there being large bullae. This will be reflected in a continuum of changes in lung density reflected by shifts in the cumulative density histogram as seen in Figure 1 .
What is critical is that efforts be made to standardize specific protocols that identify the scan parameters which best match measures across and within manufacturers. Noise levels should be held constant with CT Dose Index rather than mAs being the preferred constant because the same effective mAs across manufacturers and models can give very different doses to the subject in part because of differences in x-ray filtering across the manufacturers and models. Adjustments should be made in the protocol for subject size (body mass index) so as to further control for image noise. Kernels need to be identified which match image characteristics and sharp kernels which artificially suppress voxel density (HU) at bright/dark boarders in the image should be avoided as should algorithms which overly smooth the image. Ideally patients are imaged with the aid of spirometry to assure that the lungs are held at standardized percentages of the vital capacity, but when spirometry is impractical in large multicenter and multinational studies, subjects should be carefully coached to either total lung capacity or residual volume for assessment of emphysema-like voxels or air trapping respectively. Once a subject is entered into a study, he or she should, if at all possible, be scanned throughout the study on the same scanner. If the scanner changes at a given center, care must be taken, via use of phantoms to establish the changes in image characteristics introduced by the scanner change and the analysis software and statistical assessment must be adjusted to cope with the changes. Because it is well recognized that air outside the body can be close to -1000 HU, whereas air in the trachea can be up to 60 HU less negative, algorithms for image quantification must cope with these discrepancies.
It is clear that, as quantitative CT takes center stage in an era where old disease classifications are recognized to be inadequate for the linkage of phenotypes to genotypes and for the focused efforts to tailor disease interventions to the individual patient, radiology groups must step up to the challenge by being as rigorous in the acquisition of a chest CT scan as is the pulmonologist in the acquisition of pulmonary function tests. These challenges are being addressed through focused efforts associated with the above mentioned multicenter trials, and the Radiological Society of North America has convened a set of working groups know as the Quantitative Imaging Biomarkers Alliance to further address all of these issues.
Get Radiology Tree app to read full this article<
References
1. Sieren J., Halaweish A., Guo J., et. al.: Reconstruction kernel effects on quantitative CT airway and lung density. Am J Respir Crit Card Med 2009; 179: A3532
2. Reddy S, Barr RG, Berbaum K, et al. Quantitative evaluation of emphysema: is coronary calcium screening CT a surrogate for a full lung study? Radiological Society of North America 91st Annual Meeting, Chicago, IL: 2005, p. 638.
3. Coxson H.O., Rogers R.M., Whittall K.P., et. al.: A quantification of the lung surface area in emphysema using computed tomography. Am J Respir Crit Care Med 1999; 159: pp. 851-856.
4. Madani A., Zanen J., de Maertelaer V., et. al.: Pulmonary emphysema: objective quantification at multi-detector row CT—comparison with macroscopic and microscopic morphometry. Radiology 2006; 238: pp. 1036-1043.
5. Hsia C.C., Hyde D.M., Ochs M., et. al.: Commentaries on viewpoint: use of mean airspace chord length to assess emphysema. To be or not to be-accurate. J Appl Physiol 2008; 105: pp. 1982-1983. author reply 1986–1987