The British Journal of Radiology, 80 (2007), 414–421
Precision of computer-aided volumetry of artificial small solid pulmonary nodules in ex vivo porcine lungs 1
¨ LLER-HU¨LSBECK, H BOLTE, MD, 1C RIEDE, MD, MSc, 1S MU 1 1 1 T DREWS, M HELLER, MD and J BIEDER, MD
MD,
2
S FREITAG-WOLF,
MSc,
3
G KOHL,
MSc,
1
Department of Diagnostic Radiology, University Hospital Schleswig-Holstein Campus Kiel, Arnold-Heller Strasse 9, 24105 Kiel, 2Institut of Medical Informatics and Statistics, University Hospital Schleswig-Holstein Campus Kiel Brunswiker-Strasse 10, 24105 Kiel and 3Siemens Medical Solutions, Siemensstraße1, 91301 Forchheim, Germany
ABSTRACT. The purpose of this study was to investigate the precision of CT-based volumetric measurements of artificial small pulmonary nodules under ex vivo conditions. We implanted 322 artificial nodules in 23 inflated ex vivo porcine lungs in a dedicated chest phantom. The lungs were examined with a multislice spiral CT (20 mAs, collimation 1660.75 mm, 1 mm slice thickness, 0.7 mm increment). A commercial volumetry software package (LungCARE VA70C-W; Siemens, Erlangen, Germany) was used for volume analysis in a semi-automatic and a manual corrected mode. After imaging, the lungs were dissected to harvest the nodules for gold standard determination. The volumes of 202 solitary, solid and well-defined lesions without contact with the pleura, greater bronchi or vessels were compared with the results of volumetry. A mean nodule diameter of 8.3 mm (¡2.1 mm) was achieved. The mean relative deviation from the true lesion volume was –9.2% (¡10.6%) for semi-automatic and –0.3% (¡6.5%) for manual corrected volumetry. The subgroup of lesions from 5 mm to ,10 mm in diameter showed a mean relative deviation of –8.7% (¡10.9%) for semi-automatic volumetry and –0.3% (¡6.9%) for manually corrected volumetry. We conclude that the presented software allowed for precise volumetry of artificial nodules in ex vivo lung tissue. This result is comparable to the findings of previous in vitro studies.
A challenge of modern radiology is the increasing number of incidentally detected subcentimetre pulmonary nodules (SCPN) of unknown dignity. One reason for this development is the introduction of multislice computerized tomography (MSCT), which has led to an improvement in spatial resolution in daily routine CT examinations [1–6]. Additionally, the increasing use of computer-aided detection (CAD) software will also contribute to an increasing number of undetermined SCPN [7–9]. In a screening population, the majority (95–98%) of SCPN are reported to be benign [1, 3, 4, 10, 11]. However, lung cancer screening studies with low-dose CT showed a prevalence of asymptomatic cancers in 1.3–2.7% of a smoking population [1, 3]. Risk group screening studies also revealed that 15–24% of lung cancers are diagnosed at a size of 10 mm or less in diameter [1, 2]. Therefore, it is mandatory to reveal lesion dignity for SCPN detected in an oncological or screening setting. As malignancy is very rare in solid lesions of less than 5 mm in diameter [10], and lesions of more than 10 mm in diameter are usually surgically removed, a special interest is focused on lesion sizes from 5 to 10 mm. A general resection or biopsy of all detected SCPN in this Address correspondence to: Dr Hendrik Bolte, Department of Diagnostic Radiology, University Hospital, Schleswig-Holstein Campus Kiel, Arnold Heller Strasse 9, Kiel 24105, Germany. E-mail:
[email protected]
414
Received 9 July 2006 Revised 19 December 2006 Accepted 8 January 2007 DOI: 10.1259/bjr/23933268 ’ 2007 The British Institute of Radiology
subgroup cannot be recommended because of intervention-associated risks and for economic reasons. Therefore, the interest focuses on non-invasive diagnostic methods [12–14]. Besides conventional aspects such as lesion morphology, margins and density, new diagnostic methods such as computerized image analysis [15, 16], contrast enhancement [17, 18] or volumetric nodule growth assessment [19, 20] may prove to be valuable diagnostic criteria. The surveillance of nodule growth by the assessment of maximum lesion diameter with physical or digital callipers in repetitive CT examinations is a widespread and effective diagnostic tool. Unfortunately, especially small lesions can show significant volumetric changes while diameter changes may still not be detectable or lie within the range of measurement error [21]. Newly developed volumetric software tools have shown a high precision and good sensitivity for small volume changes [19, 22–24]. Most of these investigations were performed as in vitro studies that scanned lesions in air without an absorbing phantom or with chest CT phantoms that did not include native lung tissue [19, 20, 22, 25]. The purpose of the present study was to investigate whether, in solid, solitary lesions in contact with ex vivo lung tissue, partial volume effects resulting from adjacent small lung tissue structures would lead to a distinctly impaired precision compared with the results of studies performed under in vitro conditions. The British Journal of Radiology, June 2007
Precision of artificial pulmonary nodule volumetry in ex vivo porcine lungs
Methods and materials The ex vivo system (‘‘chest phantom’’) For this study, the authors used an existing ex vivo system for imaging porcine heart and lung explants. This system uses a copolymer container constructed to simulate a chest that holds the freshly excised, inflated lung explant of a pig by continuous evacuation of the artificial pleural space with 2–36103 Pa [26]. The original model was modified to allow for repetitive canulations and placement of multiple artificial lesions into the inflated lungs [27, 28] (Figure 1). The heart–lung explants were harvested from pigs (80– 100 kg) at a local slaughterhouse. No animals were sacrificed for the particular purpose of this study.
Preparation of pulmonary nodules The artificial lesions consisted of a fat–wax–Lipiodol’ (89% palmitin, 10% stearin and 1% Lipiodol’; Guerbet, Roissy, France) mixture that showed an absorption coefficient of approximately 50–150 Hounsfield units (HU). This material was put into a high-pressure inflation kit that was originally designed for the inflation of angioplasty balloons (Encore 26 inflation device’; Boston Scientific International, La Garenne Colombes Cedex, France) at approximately 40 ˚C and left for cooling and hardening in a water basin at 15 ˚C afterwards. Then, an extension tube and a 20G62 3/4 inch needle (Neoject; Dispomed, Gelnhausen, Germany) were fixed to the syringe for injecting the mass into the inflated lung. The needle was used to penetrate the silicon seals of the posterior chest wall to place mixture deposits at a variable depth of 3–6 cm into the lung parenchyma. During injection, the mixture liquefied under increasing pressure inside the syringe. When a flow in the extension was observed (about one turn on the piston; this equals 0.5–1.0 ml), the release button was triggered after 1–2 s, so that the flow stopped immediately. The needle was removed with a delay of 30 s in order to allow hardening of the material at room temperature [27]. In 23 lungs, mixture deposits of variable volumes were injected at 14 well-defined positions, which represented a total of 322 artificial lesions. We ordered the positions
Figure 1. Chest phantom with prepared inserted porcine lung.
The British Journal of Radiology, June 2007
by numbering the drill holes of the posterior chest wall. In the right and left lung, two lesions were injected in the posterior recessus, two in the lung base, two in the middle lung level and one in the upper lung level, respectively.
CT scanning and image analysis The prepared chest phantoms were scanned with a multidetector CT scanner (Somatom Sensation 16’; Siemens, Erlangen, Germany) using a standard volumetry protocol as recommended by the manufacturer (effective mAs 20, 120 kV, slice collimation 1660.75 mm, table feed 15 mm, reconstructed slice thickness 1 mm, reconstruction increment 0.7 mm, intermediate reconstruction kernel (B50), field of view (FOV) 350 mm, matrix 5126512 pixel, CTDIvol 1.92 mGy). Image and volume analysis was performed using the consensus of two experienced radiologists (SM-H, JB). Initially, all lesions were evaluated in a two observer consensus scheme regarding nodule size, morphology and density in a standard image setting (window width 1600 HU and centre at 2600 HU) on a commercial workstation (Wizard’, VA70C; Siemens, Erlangen, Germany) using digital callipers and regions of interest (ROI). The nodule size was defined as the longest diameter in the axial view. All nodules with good demarcation, roundish shape and a solid aspect were chosen to resemble small solid nodules such as metastases or granulomas. Lesions with poor demarcation, a ground-glass aspect or distinct draining of mixture into a bronchus, blood vessel or into the injection pathway were discarded. Lesions in direct contact with pleura or larger bronchial or vascular structures were discarded too. For volume analysis, a commercially available volumetry software (LungCARE’; Siemens, Erlangen, Germany [29]) was used. This software offers a computer read-only semi-automatic volumetry and the option to modify the subvolumes of a lesion that are included in the calculated volume. In order to create a volume of interest (VOI), a nodule is selected with a mouse click into the lesion centre. The initial structure of interest is marked by a coloured rectangle. If the observer accepts the positioning of this marker, primary (semiautomatic) volumetry is started. Alternatively, the observers could manually change the included volume by volume increase or decrease according to the calculated cut-off level. Again the included volume can be monitored by changes of the coloured overlay (Figures 2 and 3a–c). Based on this process, the investigators compared the estimated volumes of two different software operating modes with the gold standard: results of ‘‘semi-automatic volumetry’’, obtained without any observer induced post processing, and results of ‘‘manual corrected volumetry’’, including manual correction of the marked lesion volume to fit the lesion margins. As the applied software tool has been optimized for the evaluation of solid, soft tissue, density nodules of approximately spherical shape, part-solid lesions or nodules of ground-glass opacity (GGOs) cannot be evaluated correctly. 415
H Bolte, C Riedel, S Mu¨ller-Hu¨lsbeck et al
Figure 2. Graphical user interface of the LungCARE’ software showing four small screens (default setting): two axial views, of which one has a set slice thickness of 1 mm (above right) and one represents a thick sliding maximum intensity projection (MIP) of adjustable thickness (above left), a coronal view in which the thickness of the thick sliding MIP can be adjusted (below left) and one view that shows the select volume of interest (VOI, below right). To create a VOI, a nodule is selected with a mouse click. The initial structure of interest is marked by a rectangle. If the observer accepts the positioning of this marker, primary (semiautomatic) volumetry is started.
Retrieval and volume calculation of artificial nodules Initially, the specific density (r) of the fat–wax– Lipiodol’ mixture was defined by means of precision weighing of the mixture in a cylinder with diameter (d) of 30 mm and height (h) of 30 mm with a commercial laboratory scale (SartoriusH CP124S-OCE, capable of 1 mg precision). The cylinder was weighed five times consecutively at room temperature. The obtained mean 416
weight and the volume {V[p6h6(d/2)2]} were used to calculate the specific density by r5m/V, which was 0.897 g cm23. After imaging, the lungs were deflated, cooled down to 4 ˚C and dissected to harvest the nodules. The retrieval was done with knowledge of the documented nodule positions in the CT scan, which allowed for a gentle preparation. Although the lungs were collapsed, the positions still correlated well with the locations under scanning conditions. Nodules with inaccurate dissection The British Journal of Radiology, June 2007
Precision of artificial pulmonary nodule volumetry in ex vivo porcine lungs
(a)
(b)
(c)
Figure 3. Process of nodule selection and volumetric analysis: (a) marked lesion without volumetric analysis, (b) result of semiautomatic volumetry (the included subvolume is marked homogeneously white), (c) after manual corrected volumetry, all lesion parts are marked.
(i.e. damage, deformation or incomplete removal) were discarded. The retrieved nodules were weighed three times on the laboratory scale at room temperature, and the mean result was taken for further analysis. Knowing the mean weight and the specific density of the mixture (r50.897 g cm23), we calculated the reference volume (V) of each retrieved lesion (V5m/r).
Statistical analysis Only nodules that met the morphological criteria (good demarcation, solid nodule shape and no draining of mixture) and that could be dissected without any damage were used for statistical analysis. To evaluate the precision of semi-automatic and manual corrected measurements, the volumetric results had to be compared with the dissected lesion volumes. To assess measurement deviation, we calculated the relative deviation of the volumetric results from the physically measured reference volumes. In all cases, the mean value (mean), the empirical standard deviation (SD) and the 95% confidence interval (95% CI) were calculated. All results were presented separately for semi-automatic and manual corrected volumetry. As a normal distribution was given, the means and variances of semi-automatic and manual corrected volumetry were tested for significant differences with paired t-tests and F-tests. An analysis of variances (ANOVA) was performed to identify potential influence of nodule size upon the relative deviation. For all tests, a significance level of 5% was chosen.
Results Morphology Some 71.7% (n5231) of the artificial lesions could be accurately dissected. From this subgroup, 87.4% (n5202) demonstrated the morphological signs of small, solid lung nodules, which were used for volumetric analysis. The mean nodule density of this subgroup was 72.1 HU, ranging from 4 to 142 HU with a SD of ¡31.1 HU. The mean planimetric nodule diameter was 8.3 mm (¡2.1 mm), ranging normally distributed from 2.8 to 18.2 mm. The lesions were subdivided into groups of nodules of less than 5 mm (n52), 5 to less than 10 mm (n5171) and more than 10 mm (n529) in diameter. Because there were only two lesions of less than 5 mm in diameter, further statistical analyses in this group were discarded.
Overall results For the evaluation of overall deviation of semiautomatic and adapted volumetry, 202 lesions were available. The mean absolute volume for semi-automatic volumetry was 109.4 mm3 (¡40.5 mm3) and 120.5 mm3 (¡43.5 mm3) for manual corrected volumetry. The relative deviation of semi-automatic volumetry from the retrieved nodule volumes was –9.2% (¡10.6%). For manual corrected volumetry, the mean relative deviation was –0.3% (¡6.5%) (Tables 1 and 2; Figures 4–7).
Table 1. Overall results of volumetry presented with mean, standard deviation (SD), 95% confidence interval (95% CI), minimum (Min) and maximum (Max) in mm3 Volumetry
n
Semi-automatic 202 Manual corrected 202
Mean (mm3)
SD (mm3)
95% CI (mm3)
Min (mm3)
Max (mm3)
109.4 120.5
¡40.5 ¡43.5
[30.0–188.8] [35.2–205.8]
17.7 30.9
237.9 241.9
The British Journal of Radiology, June 2007
417
H Bolte, C Riedel, S Mu¨ller-Hu¨lsbeck et al Table 2. Overall relative deviation presented with mean, standard deviation (SD), 95% confidence interval (95% CI), minimum (Min) and maximum (Max) as a percentage Volumetry
n
Semi-automatic 202 Manual corrected 202
Mean (%)
SD (%)
95% CI (%)
Min (%)
Max (%)
–9.2 –0.3
¡10.6 ¡6.5
[–29.9 to 11.6] [–13.0 to 12.4]
–63.7 –32.6
32.4 43.9
In the majority of lesions (n572, 85.1%), the semiautomatically included subvolumes were smaller than the true lesion volume, and the marked subvolume has to be increased manually to fit properly with the true lesion volume (manual corrected). The different results between the semi-automatic and manual corrected method were significant (paired t-test and F-test: p,0.001). Nodule diameter showed no significant effect on the mean deviation of semiautomatic volumetry (p50.45) and manual corrected volumetry (p50.947). However, the variances were significantly affected by nodule size (F-test: p50.037 for semi-automatic and p,0.001 for manual corrected volumetry).
Results for nodule diameters 5 to less than 10 mm 171 nodules (84.0%) belonged to this subgroup. Semiautomatic volumetry showed an absolute mean of 106.8 mm3 (¡39.5 mm3) and manual corrected volumetry an absolute mean of 116.7 mm3 (¡41.6 mm3) (Tables 3 and 4). Calculating the relative deviation of volumetric results obtained from computer-aided volumetry and the retrieved nodule volumes, for semi-automatic volumetry, a mean relative deviation of –8.7% (¡10.9%) could be revealed. For corrected volumetry, the mean relative deviation was –0.3% (¡6.9%) (Tables 3 and 4, Figures 8 and 9). The different results between the semi-automatic and manual corrected methods were significant (paired t-test: p,0.001; F-test: p,0.001).
Figure 4. Scatterplot of overall results (n5202) of semiautomatic volumetry compared with dissected lesion volume. 418
Figure 5. Scatterplot of overall results (n5202) of manual corrected volumetry compared with dissected lesion volume.
Results for nodule diameters of 10 mm and more In this group, we included 29 nodules for statistical analysis (14.3%). Semi-automatic volumetry revealed a mean volume of 127.6 mm3 (¡41.64 mm3) and manual corrected volumetry a mean of 146.1 mm3 (¡45.98 mm3). The mean relative deviation of semi-automatic volumetry was –12.9% (¡7.8%). For manual corrected volumetry, the mean relative deviation was –0.3% (¡3.9%) (Tables 3 and 4). Again, the paired t-test and the F-test proved significant differences between the results of semi-automatic and adapted volumetry (both tests: p,0.001).
Figure 6. Scatterplot of overall results (n5202) for relative deviation of semi-automatic volumetry compared with dissected lesion volume.
The British Journal of Radiology, June 2007
Precision of artificial pulmonary nodule volumetry in ex vivo porcine lungs
to be known in advance, or there has to be an option of dissection and nodule retrieval after each examination. In this study, we chose the latter approach and modified a dedicated chest phantom [26] as a test system that enables the use of fresh ex vivo porcine lung tissue with simulated pulmonary nodules [27]. The presented settings enabled us to produce a large number of subcentimetre artificial pulmonary lesions that could be retrieved after CT examination. A large portion of lesions (62.7%) was usable for subsequent volumetric analysis. As the density of lung nodules is described to be of ‘‘soft tissue attenuation’’ [34, 35] and benign granulomas often show a density of more than 100 HU (because of common calcifications), we aimed to achieve a density of artificial nodules that lies in between that of malignant non-calcified lesions and benign calcified lesions (72.1¡31.1 HU). Although the injected lesions consisted of a homogeneous fat–wax–Lipiodol’ mixture, the variance of measured lesion density can be explained by partial volume effects that may be increased by contact with native lung tissue and slightly irregular lesion shape. For lesions of 5–10 mm diameter, the detected mean relative deviation was –8.7% for semi-automatic and –0.3% for manual corrected volumetry. For lesions of more than 10 mm in diameter, mean relative deviations were –12.9% and –0.3%, respectively. The results for lesions of 5–10 mm in diameter can be compared with the findings of previous experimental investigations. In an in vitro study with spherical artificial lesions, Yankelevitz et al [22] found a slightly lower relative error of volumetry of approximately 3% at a slice thickness of 1 mm. However, in that experiment, the lesions were scanned in air without an absorbing phantom and at 200 mAs tube current. Volumetry was performed with a different software tool (VISIONX; Cornell University, Ithaca, NY). Using a commercial chest CT phantom and spherical lesions, Ko et al [19] revealed a mean relative error of approximately 7.7% for lesions of 5 mm diameter at a tube current of 20 mAs and 1.25 mm slice thickness. In that study, software from the same manufacturer as in the presented study was used. Goo et al [25], using the same phantom type as Ko et al, scanning at a slice thickness of 1 mm and 255 mAs, but using different software (Rapidia, Seoul, Korea), found a mean relative error of 9.4% for lesions of 6.4 mm in diameter and of 5.4% for those 12.7 mm in diameter. The presented study could not reproduce the previously described significantly reduced precision for smaller lesions explained by the larger proportion of surface voxels affected by partial volume effects [19, 22, 25]. This may be caused by the smaller number of lesions in the subgroup .10 mm, which makes the statistical analysis susceptible for outliers. In addition, for semi-
Figure 7. Scatterplot of overall results (n5202) for relative deviation of manual corrected volumetry compared with dissected lesion volume.
Discussion The improved spatial resolution of modern routine CT and the use of CAD software products increases the number of incidentally detected small pulmonary nodules [1–9]. As a non-negligible portion of these lesions is malignant, new diagnostic algorithms to expedite assessment of lesion dignity are desirable [2, 3, 11, 30]. Computer-assisted lung nodule volumetry is likely to become a new diagnostic tool that may help to improve the characterization of subcentimetre pulmonary nodules [13, 14]. An essential prerequisite is to evaluate possible errors concerning precision, reproducibility and interobserver variability to determine a growth detection threshold. Precision of nodule volumetry has been investigated previously in different studies that predominantly used in vitro phantoms [19, 22, 25]. However, it is of great interest to investigate volumetric precision for lesions situated in lung tissue. Partial volume effects of connected tissue components may lead to more inaccurate measurements compared with the results of studies performed under in vitro conditions. The three-dimensional CT setting allowed no use of established phantoms that apply superimposed templates or nodules, as commonly used for radiographic experiments [31, 32]. Furthermore, we intended not to use an in vivo animal model [33] in order to reduce ethical problems and expenditure. Another crucial aspect of a non-in vitro setting is to know the true lesion volume. Either lesion volumes have
Table 3. Size-adapted results of volumetry presented with mean, standard deviation (SD), 95% confidence interval (95% CI), minimum (Min) and maximum (Max) in mm3
Size (mm)
Volumetry
n
Mean (mm3)
SD (mm3)
95% CI (mm3)
Min (mm3)
Max (mm3)
5 to ,10
Semi-automatic Manual corrected Semi-automatic Manual corrected
171 171 29 29
106.8 116.7 127.6 146.0
39.5 41.6 41.6 45.9
[29.4–184.2] [35.2–198.2] [46.1–209.1] [64.5–227.5]
17.7 30.8 55.9 66.9
237.8 241.8 220.7 240.6
>10
The British Journal of Radiology, June 2007
419
H Bolte, C Riedel, S Mu¨ller-Hu¨lsbeck et al Table 4. Size-adapted relative deviation presented with mean, standard deviation (SD), 95% confidence interval (95% CI), minimum (Min) and maximum (Max) as a percentage Size (mm)
Volumetry
n
Mean (%)
SD (%)
95% CI (%)
Min (%)
Max (%)
5 to ,10
Semi-automatic Manual corrected Semi-automatic Manual corrected
171 171 29 29
–8.7 –0.3 –12.9 –0.3
¡10.9 ¡6.9 ¡7.8 ¡3.9
[–30.1 [–13.8 [–28.2 [–7.9
–63.7 –32.6 –34.5 –10.7
32.4 43.9 0.4 8.9
>10
automatic volumetry, in most lesions the area indicating the included volume did not cover the whole lesion and had to be corrected manually (Figure 3a–c). This may be because the applied algorithm is optimized for SCPN. Probably for the same reason, manual corrected volumetry was superior to semi-automatic volumetry, as the measured subvolume had to be increased manually to fit properly with the presented lesion size in many cases. Several study limitations have to be discussed. The presented data were retrieved under ex vivo conditions. In vivo effects such as motion artefacts from cardiac or vascular pulsation could not be estimated. Thus, the presented findings have be very carefully transferred to the clinical situation. In this experiment, only solid and solitary nodules without contact with larger lung structures of similar density, such as blood vessels, the bronchial tree or the pleura, were investigated exclusively. Particularly for threshold-based segmentation techniques, inaccurate measurements can be expected in nodules with a connected component as the algorithm cannot differentiate between lesion and connected tissue [19]. At present, this condition poses the biggest problem in volumetry, although a current study by Kuhnigk et al [36] using a modern software algorithm showed promising results for this lesion category. Intra- and interscan repeatability represents a more important aspect of lesion volumetry than precision. An accurate repeatability is a prerequisite for the detection of volume changes in follow up measurements. The investigation of repeatability was subject to previous ex vivo studies [28]. Intra- and interobserver variability can also influence the
Figure 8. Scatterplot of relative deviation of semi-automatic volumetry compared with dissected lesion volume for lesions from 5 to 10 mm in diameter (n5171). 420
to to to to
12.7] 13.2] –2.4] 7.3]
Figure 9. Scatterplot of relative deviation of manual corrected volumetry compared with dissected lesion volume for lesions from 5 to 10 mm in diameter (n5171).
reliability of a volumetric method. As consensus reading was applied, this error could not be quantified. Finally, it has to be kept in mind that the used gold standard volumes (of the retrieved nodules) are calculated in knowledge of the specific weight of the injected fat–wax– Lipiodol’ mixture. This may represent an additional source of uncertainty as nodule retrieval after scanning could be associated with a dissection error (i.e. loss of material resulting in underestimation of the true volume). On the other hand, it can be expected that, because of the large number of examined lesions, this possible error could only increase the variation (i.e. SD and 95% CI), but the mean should be unaffected. However, the achieved mean volume deviation was comparable with the results of previous studies. The authors considered this methodical disadvantage as a ‘‘tribute’’ to the ex vivo setting, which raised complex difficulties of lesion retrieval and placement compared with scanning exact spherical phantoms in air or in vitro. It can be concluded that, compared with the results of previous in vitro studies, contact with native lung tissue does not seem to have a distinct effect on the precision of lung nodule volumetry. To achieve maximum precision, observer-based post processing is still needed. However, this may become a source of increased interscan or interobserver variability and may impair reproducibility. The dimensions of the different effects have to be known and weighed against each other. It can be expected that further developed, click point independent segmentation techniques may be able to overcome this problem with more sophisticated semi-automatic algorithms. The British Journal of Radiology, June 2007
Precision of artificial pulmonary nodule volumetry in ex vivo porcine lungs
References 1. Henschke CI, McCauley DI, Yankelevitz DF, Naidich DP, McGuinnes G, Miettinen OS, et al. Early lung cancer action project: overall design and findings from baseline screening. Lancet 1999;354:99–105. 2. Jett JR. Screening for cancer in high risk groups: current status of low dose spiral CT scanning and sputum markers. Semin Respir Crit Care Med 2000;21:385–92. 3. Diederich S, Wormanns D, Semik M, Thomas M, Lenzen H, Roos N, et al. Screening for early lung cancer with low-dose spiral CT: prevalence in 817 asymptomatic smokers. Radiology 2002;222:773–81. 4. Swensen SJ, Jett JR, Sloan JA, Midthun DE, Hartmann TE, Sykes AM, et al. Screening for lung cancer with low-dose spiral computed tomography. Am J Respir Crit Care Med 2002;165:508–13. 5. Pastorino U, Bellomi M, Landoni C, De Fiori E, Arnaldi P, Picchio M, et al. Early lung-cancer detection with spiral CT and positron emission tomography in heavy smokers: 2year results. Lancet 2003;362:593–7. 6. Fischbach F, Knollmann F, Griesshaber V, Freund T, Akkol E, Felix R. Detection of pulmonary nodules by computed tomography: improved detection rate with reduced slice thickness. Eur Radiol 2003;13:2378–83. 7. Awai K, Murao K, Ozawa A, Komi M, Hayakawa H, Hori S, et al. Pulmonary nodules at chest CT: effect of computer aided diagnosis on radiologists’ detection performance. Radiology 2004;230:347–52. 8. Lee JW, Goo JM, Lee HJ, Kim JH, Kim S, Kim YT. The potential contribution of a computer-aided detection system for lung nodule detection in multidetector row computed tomography. Invest Radiol 2004;39:649–55. 9. Wormanns D, Beyer F, Diederich S, Ludwig K, Heindel W. Diagnostic performance of a commercially available computer-aided diagnosis system for semi-automatic detection of pulmonary nodules: comparison with single and double reading. Fortschr Ro¨ntgenstr 2004;176:953–8. 10. Libby DM, Smith JP, Altorki NK, Pasmantier MW, Yankelevitz D, Henschke C. Managing the small pulmonary nodule discovered by CT. Chest 2004;125:1522–9. 11. Miller DL. Management of the subcentimeter pulmonary nodule. Semin Thorac Cardiovasc Surg 2002;14:281–5. 12. Ko JP, Naidich D. Lung nodule detection and characterization with multislice CT. Radiol Clin North Am 2003;41:575–97. 13. Wormanns D, Diederich S. Characterization of small pulmonary nodules by CT. Eur Radiol 2004;14:1380–91. 14. Wiemker R, Rogalla P, Blaffert T, Sifri D, Hay O, Shah E. Aspects of computer-aided detection (CAD) and volumetry of pulmonary nodules using multislice CT. Br J Radiol 2005;78:46–56. 15. Henschke CI, Yankelevitz DF, Mateescu I, Brettle DW, Rainey TG, Weingard FS. Neural networks for the analysis of small pulmonary nodules. Clin Imaging 1997;21:390–9. 16. McNitt-Gray MF, Hart EM, Wyckoff N, Sayre JW, Goldin JG, Aberle DR. A pattern classification approach to characterizing solitary pulmonary nodules imaged on high resolution CT: preliminary results. Med Phys 1999;26:880–8. 17. Zhang M, Kono M. Solitary pulmonary nodules: evaluation of blood flow patterns with dynamic CT. Radiology 1997;205:471–8. 18. Swensen SJ, Viggiano RW, Midthun DE, Muller NL, Sherrick A, Yamashita K, et al. Lung nodule enhancement at CT: multicenter study. Radiology 2000;214:73–80.
The British Journal of Radiology, June 2007
19. Ko JP, Rusinek H, Jacobs EL, Babb JS, Betke M, McGuinnes G, et al. Small pulmonary nodules: volume measurement at chest CT – phantom study. Radiology 2003;228:864–70. 20. Kostis WJ, Reeves A, Yankelevitz DF, Henschke CI. Threedimensional segmentation and growth rate estimation of small pulmonary nodules in helical CT images. IEEE Trans Med Imaging 2003;22:1259–74. 21. Revel MP, Bissery A, Bienvenu M, Aycard L, Lefort C, Frija G. Are two-dimensional measurements of small noncalcified pulmonary nodules reliable? Radiology 2004;231:453–8. 22. Yankelevitz DF, Reeves AP, Kostis WJ, Zhao B, Henschke CI. Small pulmonary nodules: volumetrically determined growth rates based on CT evaluation. Radiology 2000;217:251–6. 23. Takashima S, Sone S, Li F, Maruyama M, Matsushita T, Takayama F. Intermediate solitary pulmonary nodules revealed at population-based CT screening of the lung: using first follow-up diagnostic CT to differentiate benign and malignant lesions. AJR Am J Roentgenol 2003;180:1255–63. 24. Jennings SG, Winer-Muram HT, Tarver RD, Farber MO. Lung tumor growth: assessment with CT – comparison of diameter and cross-sectional area with volume measurements. Radiology 2004;231:866–71. 25. Goo JM, Tongdee T, Tongdee R, Yeo K, Hildebollt CF, Bae KT. Volumetric measurement of synthetic lung nodules with multi-detector row CT: effect of various image reconstruction parameters and segmentation thresholds on measurement precision. Radiology 2005;235:850–6. 26. Biederer J, Heller M. Artificial thorax for MR imaging studies in porcine heart–lung preparations. Radiology 2003;226:250–5. 27. Bolte H, Mu¨ller-Hu¨lsbeck S, Riedel C, Jahnke T, Inan N, Heller M, et al. Ex-vivo injection technique for implanting solid pulmonary nodules into porcine lungs for multi-slice CT studies. RoFo 2004;176:1380–4. 28. Bolte H, Riedel C, Jahnke T, Inan N, Freitag S, Kohl G, et al. Reproducibility of computer aided volumetry of artificial small pulmonary nodules in ex vivo porcine lungs. Invest Radiol 2006;41:28–35. 29. LungCare (computer program). Version Somaris/5:VA70CW. Siemens, Erlangen, Germany; 2003. 30. Munden R, Pugatch R, Liptay M, Sugarbaker DJ, Le LU. Small pulmonary lesions detected at CT: clinical importance. Radiology 1997;202:105–10. 31. Bernhardt TM, Rapp-Bernhardt U, Lenzen H, Rohl FW, Diederich S, Papke K, et al. Diagnostic performance of a flat panel detector at low tube voltage in chest radiography: a phantom study. Invest Radiol 2004;39:649–55. 32. Uffmann M, Prokop M, Eisenhuber E, Fuchsjager M, Weber M, Schaefer-Prokop C. Computed tomography and direct radiography: influence of acquisition dose on the detection of simulated lung lesions. Invest Radiol 2005;40:249–56. 33. Johnson MS, Coakley FV, Carmody TJ, Dreesen RG, Cohen MD. Technique for construction of an in-vivo model of simulated pulmonary metastases. Invest Radiol 1999;34:366. 34. Webb WR. Solitary and multiple nodules, masses, cavities and cysts. In: Webb WR, Higgins CB, editors. Thoracic imaging: pulmonary and cardiovascular radiology. Philadelphia: Lippincott Williams and Wilkins, 2005: 271– 305. 35. Swensen SJ, Viggiano RW, Midthun DE, Muller NL, Sherrick A, Yamashita K, et al. Lung nodule enhancement at CT: multicenter study. Radiology 2000;214:73–80. 36. Kuhnigk JM, Dicken V, Bornemann L, Bakai A, Wormanns D, Krass S, et al. Morphological segmentation and partial volume analysis of solid pulmonary lesions in thoracic CT scans. IEEE Trans Med Imaging 2006;25:417–37.
421
