CT-Densitometry A structure-based quantitative analysis of lung-ct in emphysema

Transcription

1 CT-Densitometry A structure-based quantitative analysis of lung-ct in emphysema R.A. Blechschmidt and U. Lörcher Department of Radiology, Deutsche Klinik für Diagnostik Aukammallee 33, Wiesbaden, Germany 1. INTRODUCTION Quantitative image analysis is useful in the diagnosis of lung-diseases, such as emphysema for example. Emphysema is a pathological dilatation of air sacs with an impaired lung function. Clinical diagnosis of emphysema is unreliable [1]. Spirometric tests are globally measured values. Combined effects of restrictive and obstructive disorders can not be differentiated. Hence computed tomography (CT) is the only method to establish the definite diagnosis in consideration of morphology and quantitative analysis [2]. CT is used to determine the exact diagnosis, location of pathological lesions and the extent of disease. The quantitative evaluation of lung-ct is a reliable basis for radiological diagnosis, in particular for follow-up. The visual assessment has a widespread interobserver variance. The current methods use the global emphysema-index (EI) to characterize the degree of pulmonary destruction. We have developed a new method of evaluation thoracic CT based on pulmonary morphology. The purpose of this study was to prove the reliability of the new method and compare it with current techniques. 2. MATERIAL AND METHODS During a 6-month period, we performed lung-ct in 20 patients with emphysema (11 male and 9 female patients, aged years; mean 60 years) and in 20 patients with normal lungfunction tests (8 male and 12 female patients, aged years; mean 54 years). CT-scans, spirometric tests and clinical data were obtained for all patients. The spirometric tests included residual volume (RV), total lung capacity (TLC) and single-breath-diffusing capacity (KCO, TCO). All scans were performed using a Somatom plus 4 (Siemens, Germany) with normal inspiration (i.e. 50% of actual vital capacity), 15 mm table feed, 5 mm slice, 140 kv, 51 mas (average) and high-resolution-algorithm (AB 82). For lung identification, densityvalues of to -200 HU were used. Preparation of evaluation For each subject we selected six slices of lung-ct with a distance of 30 mm (which corresponds to every other slice). The reference scan (the third selected slice) was done at the level of the carina. A PC-based evaluation program, written in IDL-language, automatically segmented the lung-contour and marked both lobes of the lung separately as a region of interest (ROI) [3].

2 Usual method Currently the global emphysema-index (EI) is used to characterise emphysematic destruction. The EI is defined as the quotient of the area between the limits [ HU] and the total area of the lung. The upper limit is given as between -950 HU [4] and -910 HU by other authors. New method All areas between the limits [ HU] are marked. Continuous areas will be referred to as bubbles in this text. They correspond to alveoli, bronchi or emphysema bullae. See figure 1 with two examples of CT-image and corresponding marked areas. Figure 1a: Normal lung There are many small bubbles. Figure 1b: Emphysematous lung There are many large bubbles affected by emphysema. The bubbles are classified by size as follows: [0..1 mm 2 ), [1..3 mm 2 ), [3..10 mm 2 ), [ mm 2 ), [ mm 2 ), [100.. mm 2 ]. We use the distribution of the bubbles within these classes to characterise pathological cases. For both the healthy and emphysema subjects, we calculated the mean value for each class and the global EI-value. To test for differences between healthy and emphysematous lungs, the independent samples U test was performed for each class and the EI respectively. P values less than 0.05 were considered significant. 3. RESULTS First of all we had to verify that the six selected slices with d=30 mm were representative of the whole lung. Therefore we selected six further slices shifted by 15 mm, so that the twelve slices covered virtually the entire lung. The Wilcoxon matched pairs signed rank-test was performed on all the slices in pairs, with the first slice of the original dataset matched with the first slice of the shifted dataset. The distance between the slices is d=15 mm. There was no significant difference in the distribution even with α=20 %. Figure 2: Topogramm of the thorax

3 For both methods we calculated the mean and the standard-deviation for each value in the classes and the EI-value. As there was no Gauss-normal-distribution provable for some datasets, the Mann-Whitney U Test, as the non-parametric equivalent of the t-test, was performed. See appendix for table of data. number of bullae in classes number of bullae healthy emphysema 1 [0..1) [1..3) [3..10) [10..30) [ ) [100..] classes of bullae [mm 2 ] Figure 3: Number of bullae in classes For the new method the number of marked areas in the higher classes [ mm 2 ), [ mm 2 ) and [100.. mm 2 ] showed a very distinctive separation between emphysema and normal cases. Other classes did not differ. The significance-level for different distributions was p In addition the Spearman-correlation-coefficient between the values of both methods and the spirometric tests (RV, TLC, TCO, KCO) were calculated. The results suggest that residualvolume (RV) is the most characteristic lung-function-test for emphysema. The best correlation to the RV was found for the new method in the classes [ mm 2 ), [ mm 2 ) and [100.. mm 2 ], where the significance-levels were p correlation with residualvolume (RV) correlation coefficients 1 new method 0,8 old method 0,6 0,4 0,2 0 [0..1) [1..3) [3..10) [10..30) [ ) [100..] classes of bubbles [mm 2 ] Figure 4: Correlation of both methods with the residualvolume (RV)

4 4. DISCUSSION The selected six slices were representative of the whole lung. This was essential for the validity of all results. Other workgroups have previously used three slices [4] or only two slices at in- and expiration [5]. We recommend the analysis of at least six slices because we observed relevant regional differences from apex to base of the lung. With our system analysis takes less than 5 minutes inclusive of data documentation. CT is performed with low x-ray doses of 52 mas on average in order to minimize x-ray exposition. Hence time and x-ray exposition are no reasons to reduce the number of slices. In consideration of new trends in 3D-visualisation we should use all available slices (15-20) in the future. All subjects within the normal group (n=20) were found to have normal lung-function tests and no history of pulmonary diseases. The group of 20 patients with emphysema have distinctive pathological lung-function tests (RV) and reduced gas exchange (TCO, KCO). See appendix for table of data. The global EI is correct in many cases, however we observed patients with an excessive lung destruction, but with a less expressive EI value. A more precise evaluation of CT-scans showed that areas affected by emphysema are combined with large areas of higher density, which decrease the global EI-value. The analysis of manual selected ROI s is more sensitive but on the other hand is dependent on the observer and not easily reproducible. In order to consider pulmonary morphology and to exclude the influence of the observer on the results we invented the new method of automatic evaluation of thoracic CT. The new method -number of marked areas/bubbles- is based on the conception of emphysema that air sacs become bigger and confluence to bullae. The classes [0..1 mm 2 ) and [1..3 mm 2 ) represent alveoli and thus healthy subjects will have a large number in these classes. Hence emphysema patients do not differ in these classes significantly. The classes [ mm 2 ) to [100.. mm 2 ] represent more and more bronchi, but the number of bronchi is limited by anatomical reasons. Greater values in these classes indicate bullous changes. Since the classes represent the size of the bubbles, the number of bubbles in each class allows a grading in small, middle and large-sized emphysema bullae. Model predictions and experimental results seem to fit, as the difference between emphysema and normal cases is very distinct in classes that correspond to the size of emphysema bullae. Likewise the correlation-coefficients of class-values and lung-function tests are best for this classes. A further advantage of this method is the value of the lower classes representing the gas exchanging air sacs; the better the value the better the gas exchange. The new method is reproducible and very fast when implemented on standard PC-systems. Analysis of six slices including automatic lung segmentation, display and the generation of a datafile for findingsdocumentation take less than one minute. The findings-documentation is realised with a macro in word 7.0 (Microsoft). 5. CONCLUSIONS In conclusion, the classified number of bubbles appears to be a more sensitive indicator of emphysema than the global EI. However additional studies are necessary in order to validate the new method. Emphysema-grading with CT reflects the degree of lung parenchymal damage. Hence CT-densitometry has a clinical relevance and should be performed for all thoracic CT of the lung. We recommend the new method as the base of quantitative analysis of lung-ct in emphysema.

5 REFERENCES 1. Lörcher U.:Hochauflösende CT - Lungenerkrankungen, Schnetztor-Verlag, Konstanz, Kemerink G.J., Lamers R.J., Thelissen G.R., van Engelshoven J.M.: The nonlinear partial volume effect and computed tomography densitometry of foam and lung. Medical Physics. 22(9): , Yuille, A.L., Cohen D. S., Hallinan P.W.: Feature extraction from faces using deformable templates. Proceeding on the IEEE Computer Society Conference on Computer Vision and Pattern Recognition: , Tuengerthal S., Kistenbrügger J., Kunz J., Bülzebruck K., Schulz V.: Vergleich von CT- Densitometrie und Lungenfunktion bei Emphysem. Aktuelle Radiologie 6(5): , Lamers R.J., Thelissen G.R., Kessels A.G., Wouters E.F., van Engelshoven J.M.: Chronic obstructive pulmonary disease - evaluation with spirometrically controlled CT lung densitometry. Radiology 193(1):109-13, 1994 APPENDIX Gesund 6,9 % ± 5,6 % Emphysem 21,0 % ± 13,5 % U-Test 59,5 [***] rrv 0,52 [**] Table 1: Global emphysema-index [0..1 mm 2 ) [1..3 mm 2 ) [3..10 mm 2 ) [ mm 2 ) [ mm 2 ) [100.. mm 2 ]. Gesund 7497 ± ± ± ± ± 30 7 ± 11 Emphysem ± ± ± ± ± ± 67 U-Test 107 [*] 73 [***] 60 [***] 44,5 [***] 36,5 [***] 44 [***] rrv 0,25 0,46 0,55 [**] 0,61 [***] 0,68 [***] 0,67 [***] Table 2: Number of marked areas/bubbles RV TCO KCO TLC Alter Gesund 90 ± ± ± ± 9 54 ± 10 Emphysem 142 ± ± ± ± ± 9 U-Test 18 [***] 20,5 [***] 27,5 [***] 61 - Table 3: Lung-function-tests note: significance-level [***] p [**] < p 0.01 [*] 0.01 < p 0.05 limits for bubbles to -930 HU numbers are mean ± standard deviation, n=20+20