The data set included 109 pulmonary nodules, diameter between 4 and 30 mm, obtained by CT scan of the thorax of 44 consecutive patients (twenty-four men and twenty women, mean age of 57,6 years, range 36-82) . Thirty-three of the 44 patients of the study population had solitary pulmonary nodules (SPNs) casually detected in chest X-ray or CT examinations performed because of other pathologic conditions (BPCO, suspicion of pulmonary embolism, pulmonary symptoms), with no story of cancer. We also detected secondary lesions in 9 patients in oncologic follow-up. The study included non–calcific solid nodules, as described in literature [reference 15], and pulmonary nodules with ground-glass attenuation or with central cavitation. Exclusion criteria were lesions >30 mm in diameter, lesions next to atelectasis and calcific or mixed lesions (solid-liquid).
CT scanning
CT scanning was performed using MSCT (Lightspeed VCT 64-slice GE Healthcare, Milwaukee, USA), during a single apnoea (full inspiration), and acquired from diaphragm to thoracic superior strict.
Thirty-three (33) nodules were studied using the following scanning parameters: 120/140 kVp, considering body weight; 120 mA; rotation time 0,6 s; table feed 39,37mm; slice thickness 1,25 mm; collimation 40 mm and reconstruction interval of 0,625 mm; “large” field of view (Scan-FOV) and high-spatial-resolution algorithm (Bone filter).
Seventy-six (76) nodules were studied using following scanning parameters: 120/140 kVp; 120 mA; rotation time 0,6 s; table feed 39,37mm; slice thickness 0,625mm; collimation 40mm and reconstruction interval of 0,3 mm; “large” field of view (Scan-FOV) and high-spatial-resolution algorithm (Bone filter).
Each image data set was reconstructed immediately after scanning using 3 different slice thickness (1,25mm; 2,5 mm; 5 mm), 2 different reconstruction algorithms (bone and standard) and 2 different FOV (Full-FOV and FOV=10 cm) as showed in Table1.
We also obtained a CT scan of the thorax with the same technical parameters after intravenous injection of contrast medium (150mL at a flow rate of 4mL/sec and with acquisition delay of 70s) for 93 of the 109 nodules, as follows:
18/93 nodules were studied using the same scanning parameters of the first CT scan (120 mA; Standard algorithm);
75/93 nodules were studied using automatic range of mA (300-700 mA; medium mA=343), Standard algorithm and all the same scanning parameters of the first CT scan.
Computed analysis and volumetric measurements
Images were analysed at a workstation (Advantage Windows 4.4, GE Healthcare) by using a dedicated volumetric software for the assessment of pulmonary nodules (Lung VCAR- GE Healthcare Technologies). The software automatically detects pulmonary nodules, separates them from adjacent structures, segments them and determines their volume. The volume of the pulmonary nodules was calculated on each of the reconstructions, with the different slice thickness, reconstruction kernel, FOV and for nodules acquired after iv contrast medium administration, for a total of 865 3D nodule volume measurements.
Two experienced radiologists evaluated all datasets in consensus (the first one with an experience of 3 years in the use of ALA1 software and 2 years in the use of LVCAR software, the second one with an experience of 1 year in the use of software LVCAR).
Mathematical design and statistical analysis
Multivariate analysis of variance (MANOVA) for repeated measures [reference 16,17] was applied to a sample of 109 pulmonary nodules using ten different techniques for the measurements of maximum diameter and volume. Wilks’ lambda multivariate test was used inside MANOVA.
Three different factors were considered for between-nodules analysis: margins, relation with surrounding lung parenchyma and neighbouring structures (nodule type) and structure (density). Within-nodules analysis was performed by comparing all measurement techniques. Interactions between factors and techniques were also evaluated.
MANOVA analysis of contrasts allowed the comparison of measurements with respect to their mean value (averaging technique measurements) to be examined for the selection of a subset of N techniques with the highest measurement accuracy. Accuracy was quantified through the root mean square error percentage, RMSE%. It was obtained by calculating the square root of the mean squared differences between current and averaged measurements, in percentage of measurement itself. Low RMSE% corresponds to high accuracy.
See statistics in sidebar: Fig2.
Moreover, we also analyzed the dependence of relative errors from nodule factors using the one-way analysis of variance (ANOVA) and the post-hoc test of Bonferroni for pairwise comparisons.
A significance level of 95% (p<0.05) was selected for all statistical analyses which were performed using the SPSS-10 software [reference 18].