Since image fusion between real-time US and pre-procedural CT or MR images can be difficult for less-experienced operators,
automatic image fusion has been developed by some US vendors in recent years.
Although automatic image fusion between real-time US and pre-procedural CT images has been introduced,18 it should also work with pre-procedural MR images.
This is because MR images generally have been preferred as a reference data set over CT images for fusion imaging since MR images provide higher contrast than CT images between the liver and the target lesions as well as intrahepatic vasculatures.5,19 Moreover,
unlike CT images in which CT scanning is usually performed in the end-inspiratory phase of patients,
MR images are usually acquired during the end-expiratory phase of patients.
Therefore,
the respiration status of MR images is closer to that of US images that are acquired when patients breathe shallowly.
Therefore,
the registration error of fusion caused by patients’ breathing motion is theoretically less in MR images than in CT images.
When these factors are taken into consideration,
fusion imaging between real-time US and pre-procedural MR images that is easy to use and accurate is needed for interventional procedures.
To the best of our knowledge,
this is the first comparison study to date between manual and automatic image fusion of real-time US and pre-procedural MR images.
In this study,
we presented a new manual image fusion method that consists of orientation lock and point lock.
After orientation lock—placing an US transducer in the sagittal plane to let the US system know the direction of US transducer relative to the patient’s direction—the operator was able to fuse the real-time US and MR images by just marking the corresponding internal structure within the liver on both images.
Thanks to orientation lock,
the initial point lock could be performed using any US plane including oblique planes such as intercostal scan.
Therefore,
we did not have to spend time to find the same plane between real-time US and MR images for the initial point lock.
Even with mismatched image planes of real-time US and MR images,
the orientation lock has an effect of correcting the mismatch between real-time US and MR images.
Therefore,
this kind of workflow would have resulted in time-saving image fusion in our study even in the less experienced radiologist (median,
109.0 seconds; range,
51 – 245 seconds).
Although it is not easy to compare our results with previous investigations directly due to differences in patient population,
patients’ breathing status at the time of CT/MR scanning (inspiration vs.
expiration),
and reference data set used (CT vs.
MR images),
our manual method seems to be the fastest compared to other methods reported in the previous studies where the image fusion time ranged from 3.7 to 30 minutes.
5,9,15
In this study,
we also introduced a new automatic image fusion between real-time US and pre-procedural MR images by sweeping the liver with an US transducer and the automatic image fusion was feasible in 95 % (19/20,
hit rate: 95%) of patients.
Although it tended to take longer time than manual image fusion,
the automatic image fusion was completed within acceptable time even with the less experienced radiologist (median,
163.0 seconds; range,
83 – 303 seconds).
Given that some operators rarely perform US-guided interventional procedures and would be not familiar with fusion imaging,
our automatic image fusion is likely to help these operators or beginners.
In our study,
the registration error of initial image fusion immediately after sweeping the liver was 25 mm (range,
3.13 – 76.94) in version 1.1 and 16.63 mm (range,
5.38 – 26.62) in version 1.2,
respectively and thus,
acceptable image fusion could be achieved by performing several point locks.
If a patient has good sonographic window for the right liver,
our automatic image fusion would help operators who are not familiar with manual image fusion.
However,
it should be pointed out that automatic image fusion using the sweeping manner may not always yield desirable results as demonstrated in our study.
Given that the patients with poor sonographic window for the liver were excluded in our study,
the hit rate using this kind of sweeping manner might have been overestimated in our study.
Moreover,
like manual image fusion,
automatic image fusion also required additional point locks between real-time US image and its corresponding MR image after initial image fusion to refine the initial registration.
This process may be cumbersome,
especially for less-experienced radiologist,
which explains why the less-experienced radiologist took longer than the expert to conduct image fusion with automatic method while the time required for initial image fusion and the number of point lock needed after sweeping the liver was not different between the two radiologists.
Hence,
more robust automatic image fusion that is easy to use and does not affected by sonographic window is needed.
In terms of the time required for image fusion,
it was not statistically different between manual and automatic image fusion in the expert radiologist.
However,
it took shorter in the manual method than in the automatic method in the less-experienced radiologist (p = 0.038).
It can be explained by the following reasons.
First,
through training sessions with more than 10 cases,
the radiologist may have already been familiar with manual image fusion of this system.
In addition,
the less-experienced radiologist may be on a steep learning curve in a short time during the study period,
since our study was performed within a relatively short study period.
In terms of registration error of image fusion,
it was not significantly different between manual and automatic method in both the less-experienced and expert radiologist.
The median value of registration error was 6.40 mm (range,
2.08 – 21.19 mm) in the less-experienced radiologist with automatic method.
Although it is difficult to directly compare the results of our study with those of previous investigations,
our results seems to be better than previous studies where its mean value ranged from 8.1 mm to 19.1 mm.
18,20 Since we used MR images,
not CT images as a reference data set and 3D volume data was obtained at the end-expiratory phase,
not at the end-inspiratory phase of patients,
the registration error seems to be not high.
In conclusion,
the registration error was not different between the manual and automatic methods in both the expert and less-experienced radiologists.
The manual image fusion took shorter time than the automatic image fusion for the less-experienced radiologist while it was not significantly different for the expert radiologist.