Registration Error of Image Fusion
Manual image fusion was successful for all 20 patients whereas automatic image fusion performed by the less-experienced radiologist was failed for one case due to improper feature extraction by the system.
The details of image fusion between manual and automatic methods are summarized in Table 1.
The quality of image fusion was not different between manual (Fig. 4) and automatic methods (Fig. 5) in both the expert and less-experienced radiologists (p = 1.000 and p = 0.317,
respectively).
The number of point lock required for image fusion was not also different between manual and automatic methods in both radiologists (p = 0.435 and p = 0.096,
respectively).
In terms of registration error after image fusion,
it was not different between the two methods in both radiologists (p = 0.575,
p = 0.515,
respectively).
|
|
Manual method
|
Automatic method
|
P-value*
|
Expert
|
Quality of image fusion
|
4 (3-4)
|
4 (3-4)
|
1.000
|
Number of point lock
|
2.0 (1 – 4)
|
2.0 (1 – 3)
|
0.435
|
Registration error (mm)
|
3.18 (1.00 – 6.74)
|
3.68 (1.83 – 5.17)
|
0.575
|
Time required for image fusion (seconds)
|
66.5 (50 – 145)
|
83.0 (46 – 101)
|
0.720
|
Less-experienced radiologist
|
Quality of image fusion
|
4 (2-4)
|
4 (2-4)
|
0.317
|
Number of point lock
|
2.0 (1 – 3)
|
2.0 (1 – 5)
|
0.096
|
Registration error (mm)
|
4.92 (3.29 – 22.03)
|
6.40 (2.08 – 21.19)
|
0.515
|
Time required for image fusion (seconds)
|
109.0 (51 – 245)
|
163.0 (83 – 303)
|
0.038
|
Table 1: Comparison between manual and automatic image fusion according to the radiologists.
Data are median value and data in parentheses indicate range.
* Wilcoxon signed-rank test was used.
Regarding automatic image fusion,
version 1.1 was used for the former 12 patients (n = 5 for the expert and n = 7 for less-experienced radiologist,
respectively) and version 1.2 for the latter 8 patients (n = 5 for the expert and n = 3 for the less-experienced radiologist,
respectively).
The registration error of initial image fusion immediately after sweeping the liver with an US transducer was smaller in version 1.2 (median,
16.63 mm; range,
5.38 – 26.62) than in version 1.1 (median,
25 mm; range,
3.13 – 76.94 mm).
However,
it did not reach statistical significance (p = 0.242).
Time Required for Image Fusion
Time required for manual image fusion tended to be shorter than time required for automatic image fusion in both the expert and less-experienced radiologist.
Although it was not significantly different in the expert radiologist,
it showed difference in the less-experienced radiologist (p = 0.720 and p = 0.038,
respectively).
However,
in terms of automatic image fusion,
time required for initial image fusion immediately after sweeping the liver was not different between the expert and the less-experienced radiologist (median,
40 seconds; range,
32 – 55 seconds vs.
median,
46 seconds; range,
24 – 73 seconds; p = 0.093).
Comparison between the Expert and Less-Experienced Radiologist
Table 2 shows the comparison of image fusion between the expert and less-experienced radiologist.
The quality of image fusion was not different between the expert and less experienced radiologist in both manual (p = 0.726) and automatic method (p = 0.764).
The number of point lock required for image fusion was also not statistically different between the two radiologists in both manual (p = 0.776) and automatic method (p = 0.156).
However,
the registration error of image fusion was significantly smaller in the expert than in the less-experienced radiologist in both manual (p = 0.013) and automatic method (p = 0.014).
The time required for image fusion was significantly shorter in the expert than in the less-experienced radiologist in both manual (p = 0.034) and automatic method (p = 0.001).
|
Method for image fusion
|
Expert
|
Less-experienced radiologist
|
P-value*
|
Quality of image fusion
|
Manual method
|
4 (3-4)
|
4 (2-4)
|
0.726
|
Automatic method
|
4 (3-4)
|
4 (2-4)
|
0.764
|
Number of point lock
|
Manual method
|
2.0 (1 – 4)
|
2.0 (1 – 3)
|
0.776
|
Automatic method
|
2.0 (1 – 3)
|
2.0 (1 – 5)
|
0.156
|
Registration error (mm)
|
Manual method
|
3.18 (1.00 – 6.74)
|
4.92 (3.29 –22.03)
|
0.013
|
Automatic method
|
3.68 (1.83 – 5.17)
|
6.40 (2.08 –21.19)
|
0.014
|
Time required for image fusion (seconds)
|
Manual method
|
66.5 (50 – 145)
|
109.0 (51 – 245 )
|
0.034
|
Automatic method
|
83.0 (46 – 101)
|
163.0 (83 –303)
|
0.001
|
Table 2: Comparison of image fusion between the expert and less-experienced radiologist according to fusion method.
Data are median value and data in parentheses indicate range.
* Mann-Whitney U test was used.