Abstract
Although computer assisted volumetric quantification of human maxillary sinuses is commonly used to measure volumetric changes during life, reliability data for this procedure are lacking. The objective of this retrospective study is to test a semi-automatic virtual volumetric analysis technique on 36 CT scans of human maxillary sinuses. Three examiners with different clinical experience performed all measurements in three replicates. As principle of proof, the technique was examined on 12 phantoms with known volumes. The validation of the method revealed that the mean relative error was 0.364%. For the retrospective volumetric measurements from maxillary sinuses the intra- and inter-examiner agreement was quantified using appropriate intraclass correlation coefficients (ICC 1, k and ICC 2, k ) and the Bland–Altman analysis. ICC values ranging from 0.997 to 0.999 indicate almost perfect agreement for intra- and inter-examiner data. The Bland–Altman analysis demonstrated good intra- as well as inter-examiner agreement for the two proficient examiners and a lack of agreement for the untrained examiner. It can be concluded that this measurement procedure using CT scans could be strongly recommended for clinical application to determine the volume of human maxillary sinuses reliably.
The bone-encompassed luminal anatomy of the paired maxillary sinuses has proven to be a complex quadrangular pyramid, with the nasal wall as its base and the apex pointing toward the zygomatic process . For morphometric studies of maxillary sinuses, invasive methods such as injection of various materials or serial sections of cadavers have been superseded by virtual reality techniques. Plain two-dimensional (2D) radiography is unsuitable for volume analysis of maxillary sinuses . With the introduction of computer-assisted tomographic (CT) imaging in 1973 , three-dimensional (3D) anatomical information has become available . As real values cannot be obtained directly from living patients CT datasets are the only clinically available information on maxillary sinus size. Until now, one of the main obstacles has been a lack of information on the influence of instrumental, physical and human limitations, which could cause measurements to deviate from the ‘true’ value . Validation studies to determine the experimental uncertainty are crucial in the assessment of volumetric analysis techniques . An advantage of CT datasets over invasive methods is that repeated measurements can be performed in a ‘virtual patient’ without destruction of the object that is to be quantified .
Only a few groups have tested volumetric analysis techniques with a phantom model . Most CT measurement studies only determined the metric variables of the maxillary sinus and correlated these data to age, race, body height, cranial morphology and to the volume of other paranasal sinuses . None of the studies give data on intra- and inter-examiner agreement. With increasing requirements for objective measurement procedures it should be of particular interest for clinicians and researchers to ensure that the experimental uncertainty of CT-based volume determination is known. Potential applications of this method lie in studies including surveys of maxillary sinus volumes in larger populations and monitoring the volume change after intentional manipulation in the maxillary sinus, for example internal sinus augmentation or extraction of teeth associated with the antrum.
The objective of this retrospective study was to evaluate the applicability of a virtual analysis technique to quantify human maxillary sinus volumes. Another aim was to test the intra- and inter-examiner reproducibility of this technique for the first time.
Materials and methods
A validation of the method with well characterized phantoms was conducted to test the experimental procedure in advance. The geometrically complex anatomy of the maxillary sinuses was mimicked by moulding (Impregum™, 3M Espe, Seefeld, Germany) two different rubber ducks’ heads ( Fig. 1 ). Target volumes of 13, 17 and 21 cm 3 each were prepared using ‘class AS’ one-mark pipettes, (Brand GmbH, Wertheim, Germany) and poured into each of the two moulds. The solution contained 100 ml iopromide 300 mg/ml (Ultravist ® , Schering AG, Berlin, Germany) diluted in 100 ml tap water. The phantom set comprises 12 different simulated maxillary sinuses.
The clinical part of this investigation was conducted as a retrospective study. CT data acquisition and volume determination were performed as described below. The protocol was authorized by the Ethics Committee of the Medical University of Graz, Austria (Nr. 19-018ex07/08). Sample size calculation was based on a preliminary series of 10 maxillary sinuses (60 readings). The standard deviation between three observers was determined using a paired t -test of equivalence of means, taking half the standard deviation of 5.000 as the limit for the equivalence-difference, with alpha set at 0.05 and power at 80%. When the expected mean difference is 0.200, the minimal number of samples should comprise 31 maxillary sinuses. CT datasets of patients who had CT scans for diagnostic reasons in the ear nose and throat (ENT) region between 2003 and 2007 were retrieved from the database of the Department of Radiology, Medical University of Graz. Selection criteria were that CT data acquisition was performed exclusively by spiral CT technology (Somatom ® plus 4 CT Scanner, Siemens AG, Medical Solutions, Erlangen, Germany) using 120 kVp and 150 mA. Further inclusion criteria were that all patients were male, aged 20–30 years at the time of the CT scan, with complete dentition (without consideration of wisdom teeth). All subjects with evidence of sinus pathology such as inflammation, cysts or tumour mass within or adjacent to the maxillary antrum as well as a history of paranasal sinus surgery or maxillofacial trauma were excluded from the study. If one of the two maxillary sinuses was scanned incompletely, that CT dataset was discarded. All included CT scans were anonymized. The volume of the maxillary sinus was defined as the bone-encompassed space comparable to the cavity in a dry skull.
All volume determinations are based on semi-automated computer calculations using commercial programs and equipment. The original CT datasets were transferred to an independent workstation (Ultra 10-workstation, Sun Microsystems, Santa Clara, CA, USA) in the DICOM format and evaluated in random order by one undergraduate dental student (E1), one well experienced radiologist (E2) and one well experienced oral surgeon (E3). Specific tools of the commercial graphics program (Somaris ® Sienet Magic View 1000 ® , VB32B, Siemens AG, Medical Solutions, Erlangen, Germany) can generate 2D and 3D reconstructions for image processing and volume calculation. The 3D region of interest of each sample was defined by using the ‘draw’ function to trace the perimeter of the radiopaque liquid or the bony boundary of the maxillary sinuses on each consecutive axial CT slice displayed on the computer monitor. Appropriate density threshold coefficients, expressed in Hounsfield units (HU), were selected to exclude impression material or bony structures as well as dental roots (HU −1000 to +300). Thus, in the 3D template only radiopaque liquid or air and soft tissue (mucous membrane) were extracted by the volume rendering function of the program. The volume of the region of interest was calculated automatically.
All examiners were trained to use this standardized study procedure and the software tools. They performed measurements in a darkened room independently of each other and blinded to previous readings. Each measurement cycle was performed three times with a minimum interval of 2 weeks between sessions to minimize personal memory effects.
Data were analysed on a personal computer using Excel 2000 (Microsoft Corporation, Redmond, WA, USA), nQuery 5.0 (Statsol, Cork, Ireland) and SPSS 14.0 (SPSS Inc., Chicago, IL, USA) on a Microsoft ® Windows ® XP platform. The volumes are expressed as cm 3 (median and range). The experimental uncertainty between the observers’ measurements and their target values was represented as absolute error in cm 3 and relative error in percentage.
The intra- and inter-examiner data were tested by an appropriate intraclass correlation coefficient (ICC) to obtain the degree of agreement. Following S hrout & F leiss for the intra-examiner agreement, the ICC 1, k (one-way random model) was used. For the inter-examiner agreement in a two-way random effects layout, the ICC 2, k equation, was selected. The dimensionless values for the ICCs range from 0 to 1, where 0 means no agreement and 1 total agreement. In accordance with L andis & K och , ICC values were classified on a six-point scale as poor (0.00), slight (0.01–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80) and almost perfect (0.81–1.00) agreement . Additionally, the repeatability and agreement of measurements were assessed by the method described by B land & A ltman , consisting of ˉd
d ¯
as the mean difference; SE of ˉd
d ¯
as the standard error of the mean differences; 95% CI for ˉd
d ¯
as the 95% confidence interval for the mean of differences and SD Diff as the standard deviation of the differences and 95% limits of agreement as ˉd±2SDDiff
d ¯ ± 2 S D Diff
.
Results
The experimental uncertainties between the target phantom volumes of 13, 17 and 21 cm 3 and the calculated volumes were less than 0.5% for all examiners. Table 1 presents the number of measurements, the calculated volumes, and the absolute and relative error. The agreement plot of observations versus the three target values for the phantom models is shown in Fig. 2 .
| Target volume | k | Calculated volume/cm 3 (±SD) | Absolute error/cm 3 (±SD) | Relative error/% |
|---|---|---|---|---|
| 13 ml | 18 | 12.938 (±0.050) | 0.067 (±0.041) | 0.386 |
| 17 ml | 18 | 16.926 (±0.039) | 0.074 (±0.039) | 0.230 |
| 21 ml | 18 | 20.932 (±0.100) | 0.092 (±0.077) | 0.477 |
| Average | 54 | N/A | 0.078 (±0.055) | 0.364 |
To assess the intra- and inter-examiner agreement, 324 readings were recorded. The time required to segment the volume of one maxillary sinus varied from 17 to 22 min (19.34 ± 3.21 min) depending on the number of slices. The current study with a sample size of 36 maxillary sinuses revealed an average maxillary sinus volume of 21.99 cm 3 (±4.34 cm 3 SD, min 13.22, max 29.24) in 20–30-year-old male subjects. The ICC coefficient for intra- and inter-examiner reliability indicates almost perfect agreement ( Table 2 ). The Bland–Altman analysis estimates for the intra-examiner repeatability a one-sided confidence limit of 1.069 cm 3 or less ( Table 2 ). As the assessment of the inter-examiner agreement indicated that there is a systematic bias between examiner 1 (E1) and both examiners 2 (E2) and 3 (E3), as in the 95% CI for ˉd
d ¯
, zero was not included between the lower and the upper limit. There was no bias between examiners E2 and E3.
