Phantom object data

After careful gradient calibrations, the proposed segmentation technique for volume quantization was in a first phase checked on images obtained from water filled phantom objects. Although segmentation in this case is obvious, this test was performed to reveal possible biases in the volume quantization. The object volumes were chosen in the range of mouse brain volume: 156, 316, 459, and 756 mm³. Each phantom volume was quantized three times: the 3D data were presented to the user by slicing in the x-, y-, and z-direction.
Table 9.1 shows the true volumes of the phantom objects, along with the measured volumes. The measured volumes reveal a systematic quantization error of about 2%, while reproducibility tests revealed an intrinsic segmentation error of less than 1%, which indicates that the limiting factor in the volume quantization accuracy is due to the limited accuracy of the gradient calibration. As segmentation of our phantom object was a very simple problem, it is believed that the segmentation procedure itself was extremely accurate. Furthermore, segmentation results of the phantom objects were statistically tested on their independence of slice direction. Thereby, ANOVA (analysis of variance) was applied to the segmentation results. It was found that the null hypothesis of equal volume means (between and within the slice directions) could not be rejected at a significance level of 0.05.

Table: Accuracy of phantom volume quantization

Real object	Volume 1	Volume 2	Volume 3	Volume quantization
volumes	x-slicing	y-slicing	z-slicing	error
(mm3)	(mm3)	(mm3)	(mm3)	(%)
156	163	159	160	1.7
316	322	322	320	0.8
459	469	468	469	0.5
736	753	753	752	0.4