Reviews
Monday 19 September 2005
Summary:
The authors seek to apply Thirlon’s Demons deformable registration method to align MR scans taken as contrast perfuses through an organ. Contrast perfusion curves at anatomic points are analyzed to detect and localize tumors. The concept is that inter-scan deformations occur (one demonstration application is breast MR) and such deformations will confound the computation of perfusion corresponding anatomic points.
The research area is well motivated in the paper. Excellent experiments are conducted. This is a very good research paper.
The paper will ideally motivate others to apply open-source software to their clinical research.
This paper has limited benefit to the open source community. The Demons method and the use of histogram matching is already detailed in the Insight Software Guide.
Hypothesis:
Open-source software (ITK) can be applied to register MR image sequences that contain deformations.
Evidence:
The evaluations are excellent. Some clinical issues should perhaps be considered, but such issues are not the focus of this conference/journal.
Open Science:
The paper makes good used of open source, and it details many of the parameters involved in the algorithms used.
The data used in the paper are not made publicly available.
The software developed for the paper is not make publicly available. The application is summarized by a pipeline / flow chart, but the code to implement that pipeline is needed to fully understand the parameters of the application. How does it differ from the work presented in the Insight Software Guide?
Reproducibility:
There are parameters to histogram matching that are not given.
There are parameters to the demons method that are not given.
Experimentation would probably be required to duplicate these results.
Requests for additional information from authors:
This work has great potential, but its impact on the open-source community is limited without source code or implementation details. This paper would be an excellent demonstration of the clinical potential of open-source, if it provided the details/code needed for others to replicate the work. Perhaps, at a minimum, a simple text document that details the various parameter settings could also be uploaded? Ideally, source code would be made available.
Vincent Magnotta
Sunday 7 August 2005
It is clear that dynamic contrast MR imaging would benefit from image co-registration. Based on the image acquisition utilized in this paper, it is unclear if a nonlinear image registration is better than a rigid registration. While not described in sufficient detail in the paper, it appears that the dynamic contrast imaging has an image acquisition period of approximately 30 seconds. This is sufficient time to average several cardiac and respiratory cycles into the data causing blurring in the data as compared to non rigid motion as would be the case when using snapshot based imaging approaches. It is also unclear in the paper why histogram matching was used in comparison to a mutual information metric. The contrast changes in the area of the lesion are large (enhancement of 3 times) as shown in Figure 1. It is unclear if simple histogram matching is sufficient to allow the Thirion demon registration to work well in areas where there is a large change in contrast between the images.
