[Impact of technical and morphological factors on the precision of software-based MR tumor volumetry: a phantom...

To investigate the impact of technical and morphological parameters on the precision of software-based MR tumor volumetry (SBV) in an in-vitro experimental setting.

Tumor models were formed from a silicone compound in three different sizes with a max. diameter < 2 cm (small), 2 - 4 cm (middle), and > 4 cm (large). For each size a spherical, an elliptic and an irregular shaped model was produced. The true volume of the tumor models was established by water displacement. Tumor models were examined with a high-field MRI (TRIO, 3 Tesla, Siemens) with T 2-weighted sequences under optimized contrast conditions. Slice thickness was 1, 3 and 5 mm. The volume of the tumor models was then calculated using (1). manually driven volumetric software (SBVmanual) and (2). automatic volumetric software (SBVauto). The influence of the following parameters on the precision of SBV was analyzed: Size and shape of the tumor models, manual/automatic SBV, segmentation technique and slice thickness.

In general, SBVauto measurements showed less deviation than measurements with SBVmanual (p < 0.01). However, both methods depended significantly on morphologic factors, especially on tumor size. In small tumor models, the volume was strongly underestimated by -36.2 +/- 27.8 % (SBVmanual) and -33.1 +/- 8.6 % (SBVauto), whereas the deviation for large tumor models was only 2.0 +/- 14.7 % (SBVmanual) and 3.0 +/- 2.3 % (SBVauto; p < 0.01). The deviation of measurements increased from the "spherical" to the "irregular" shape by 9.5 % (SBVmanual) and 10.7 % (SBVauto). In addition, SBVmanual depended on technical factors. Using a "minimal" segmentation technique (e. g. excluding partial volume effects), volumes were underestimated in all cases, whereas volumes of middle and large tumor models were slightly overestimated when using a "maximum" segmentation technique (e. g. including partial volume effects; p = 0.01). Deviation of SBVmanual increased with slice thickness from 15.9 +/- 12.7 % (1 mm slices) to 27.1 +/- 21.3 % (5 mm-slices).

In general, SBVauto measurements yielded smaller deviations than SBVmanual. However, both methods showed major inaccuracy in the volumetric estimation of small and irregular shaped tumor models, thus the tumor volumetry of these tumors has to be considered inappropriate for clinical practice. Moreover, the exactness of SBVmanual depended significantly on segmentation technique and slice thickness.