Capturas AIPP
1.6.2 Biología de las principales especies
1.6.2.3 Patagonotothen ramsayi
3.5.4.1
Experiments
To evaluate the performance of this fiducial system, we fabricated a representative example embodiment of it. The main body of this example fiducial frame made by 3D
inner diameter of 20 mm and fiducial tube diameter of d=3 mm. Interior tubular fiducials were filled with high MRI contrast gelatin or gadolinium fluid (MR-Spots, Beekley Crop., Bristol, CT) and provided ample tracking spots in the transverse image to guarantee detection accuracy and stability of image analysis algorithm.
To evaluate its accuracy with real MR images, several groups of images were taken with different orientations. A Philips 3T MRI scanner was used and all images were acquired with following scanning parameter: TR = 3000 ms; TE = 90 ms; flip angle =
90◦; slice thickness = 2 mm; pixel spacing = 0.5 mm × 0.5 mm; FOV = 80 mm × 80
mm; and pixel size = 160 × 160. The experimental groups were successively scanned along depth axis zf with fixed step length but in variables on both vertical axis xf and
horizontal axis yf. We also successively set scanning planes relative to CHIC fiducial
frame at various tilt angles and twist angles to obtain a series of transverse images to evaluate the accuracy of omnidirectional rotation, see fig. 3.21(a). The alteration of tilt angle after twist will lead to both elevation angle and direction angle change: elevation and direction angle are equal to tilt angle at the same time. There were 5 groups for tilt angle and twist angle respectively (tilt angle at 0◦, 10◦, 20◦, 30◦, 40◦
and twist angle at 0◦, 5◦, 10◦, 15◦, 20◦) forming 25 groups of different combination of
tilt angle and twist angle in total. Furthermore, two identical CHIC fiducial frames were connected in series by a 50 mm long concentric connector to make an internal contrast within a group and also test expandability for long discrete measurement application along depth. The stepping rotation of both twist angles and tilt angles
were achieved by MRI scanner itself which was much precisely than moving fiducial frames manually. Fig. 3.21(b) is the photo of the real setup of the experiment. This experiment was also using a MRI head coil to enhance imaging definition while the central axis of two concentric fiducial frames was placed along the central axis of head coil. And the phantom container was the place where we put aqueous contrast medium for MRI scanner getting a proper imaging window of contrast ratio which was close to human tissue. By using image data provided by 25 groups, registration and tracking performance was evaluated by two aspects after: translation and rotation.
3.5.4.2
Accuracy Assessment
1. Evaluation of translation
Ten groups that include five groups of different twist angles and other five groups of different tilt angles were selected. Ten sequential slices were picked from each group to evaluate the accuracy of translation tracking. The relative positions and angles were known precisely from the MR image acquisition parameters, and that this information was used as a ground truth to assess accuracy buy not utilized by the tracking algorithm. The 2D central point of each transverse image should be unchangeable after registration and only transverse depth varies. The ground truth of this accuracy analysis is the location of the MRI scan-plane which is pre-defined and with fixed increments between images. As shown in Fig. 3.22, the RMS error of x0 , y0 and z0 is 0.074 mm, 0.227 mm and 0.270 mm respectively. Specifically, the
MRI Head Coil Phantom Container Two Fiducials in Series 3T MRI Scanner (a) (b)
Figure 3.21: (a) Experiment schematic. Two fiducials being connected in series by a 50 mm long concentric connector were placed on a phantom container. (b) The photo of the real installment of the whole experiment setting. This experiment used the MRI head coil inside 3T main coil to enhance imaging definition.
detection of z0 changing along central axis were presented in Fig. 3.23 separately.
It has a RMS error of 0.270 mm and standard deviation of 0.275 mm. The overall translational RMS error is 0.208 mm and standard deviation is 0.241 mm for totally 300 samples. Since the pixel size is 0.5 mm × 0.5 mm, our results here achieved sub-pixel accuracy which is ideal.
0 10 20 30 40 50 60 70 80 90 100 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Error of Translation DOF
Image Number
Error (m
m
)
x0 y0 z0Figure 3.22: RMS error measurement results of translation DOF from 100 slices under different twist and tilt angles. The RMS error of x0 , y0 and z0 is 0.074 mm, 0.227 mm and 0.270 mm respectively.
Results of the rotation accuracy study are shown in Fig. 3.24. The detection of successive twist angles (ϕ), elevation angles (α) and direction angles (η) are listed in plot (a), (b) and (c). In each case, 50 slices were selected from five control groups of tilt or twist and the true value of corresponding step length was also superimposed into the same plot. Similar to the previous evaluation of position, the ground truth of this rotation accuracy analysis is the pre-defined orientation of the MRI scan-plane. The RMS error of twist, elevation and direction angle is 0.426◦, 0.379◦ and 0.465◦
respectively. The overall angular RMS error is 0.425◦ and standard deviation is 0.524◦
for totally 150 samples.
0 10 20 30 40 50 60 70 80 90 100 -2 0 2 4 6 8 10 12 14 16 18 20
Depth
Image Number
De
p
th
/mm
Scan plane depth Calculated depth
Figure 3.23: Detection of motion in depth with 2 mm regular step distance along z axis with RMS error of 0.270 mm and standard deviation of 0.275 mm.