Intra- and interobserver variability of thyroid volume measurements in healthy adults by 2D versus 3D ultrasound

Thyroid volume measurement by ultrasonography (US) is essential in numerous clinical diagnostic and therapeutic fields. While known to be limited, the accuracy and precision of two-dimensional (2D) US thyroid volume measurement have not been thoroughly characterized.

OBJECTIVE

We sought to assess the intra- and interobserver variability, accuracy and precision of thyroid volume determination by conventional 2D US in healthy adults using reference volumes determined by three-dimensional (3D) US. Design,

METHODS

In a prospective blinded trial, thyroid volumes of ten volunteers were determined repeatedly by nine experienced sonographers using conventional 2D US (ellipsoid model). The values obtained were statistically compared to the so-called true volumes determined by 3D US (multiplanar approximation), the so-called gold standard, to estimate systematic errors and relative deviations of individual observers.

RESULTS

The standard error of measurement (SEM) for one observer and successive measurements (intraobserver variability), was 14%, and for different observers and repeated measurements (interobserver variability), 17%. The minimum relative thyroid volume change significantly different at the 95% level was 39% for the same observer and 46% for different observers. Regarding accuracy, the mean value of the differences showed a significant thyroid volume overestimation (17%, p < 0.01) by 2D relative to 3D US.

CONCLUSION

2D US is appropriate for routine thyroid volumetry. Nevertheless, the so-called human factor (random error) should be kept in mind and correction is needed for methodical bias (systematic error). Further efforts are required to improve the accuracy and precision of 2D US thyroid volumetry by optimizing the underlying geometrical modeling or by the application of 3D US.

A novel thyroid phantom for ultrasound volumetry: determination of intraobserver and interobserver variability

A novel thyroid ultrasound phantom with tissue-equivalent characteristics was designed consisting of two lobes with three lesions each. One set of lesions is manufactured with a -5 dB echo difference to the surrounding tissue, the other with -10 dB. This phantom was used as a standardized measuring object for reproducibility of two-dimensional and three-dimensional ultrasound volumetry and for an interobserver and intraobserver variability study. For the variability study, nine experienced physicians scanned all specimen three times. Each time the volumes were calculated using the ellipsoid method. A three-dimensional ultrasound scan of each specimen was performed to evaluate all volumes by multiplanar volume approximation. The results of these volume data were compared to the known true volumes. The interobserver variability ranged from -13.4% to 11.9% (median, 0.7%); the intraobserver variability from -9.1% to 16.4% (median, 3.6%). The systematic error as calculated from the total mean of all specimens is 0.5% for the interobserver variability and 4.1% for the intraobserver variability. The phantom can be used for training purposes, to improve the skills of the examining physicians by simulating real thyroid morphology, to provide a standardized reference object for long-term quality control of conventional ultrasound scanners, and the determination of the accuracy and reproducibility of volumetry using three-dimensional ultrasound systems.

The use of three-dimensional ultrasound for thyroid volumetry

Conventional two-dimensional (2-D) ultrasound is the standard method for the investigation of thyroid morphology. Volume calculations need model assumptions and are observer dependent. The present study performed with a commercially available three-dimensional (3-D) system Freescan added to a conventional ultrasound scanner compares the accuracy of conventional thyroid volumetry to several methods of 3-D volume determination. In vitro measurements were performed on thyroid phantoms with known volumes. The standard deviation of the normalized differences was 8.0% (3-D segmentation) and 10.5% (conventional). For the accuracy of volume determination in human thyroids we performed a postmortem study. The thyroid volume was calculated conventionally by the ellipsoid model and by two 3-D methods (segmentation and the newly developed multiplanar volume approximation). The reference volume was determined after resection by submersion. The standard deviation of the normalized differences was 26.9% for the conventional method, 9.7% for 3-D segmentation and 11.5% for the multiplanar volume approximation, showing significant better results for both 3-D methods and no significant difference between the 3-D methods. The 3-D system, therefore, achieves a better accuracy for thyroid volumetry than the conventional volumetry using planar images. In addition, the 3-D images are stored electronically and can be used for follow-up studies.