ITK-based Registration of Large Images from Light Microscopy: A Biomedical Application

Please use this identifier to cite or link to this publication: http://hdl.handle.net/1926/23
Inactivation of the retinoblastoma gene in mouse embryos results
in morphological changes in the placenta, which has been shown to
affect fetal survivability. The construction of a 3D virtual
placenta aids in accurately quantifying structural changes using
image analysis. The placenta dataset consisted of 786 images
totaling 550 GB in size, which were registered into a volumetric
dataset using ITK's registration framework. The registration
process faces many challenges arising from the large image sizes,
damages during sectioning, staining gradients both within and
across sections, and background noise leading to local solutions.
In this work, we implement a rigorous ITK-based preprocessing
pipeline for removing noise and employ a novel 2-level
optimization strategy for enhanced registration in ITK. We provide
3D visualizations and numerical results to demonstrate our
improvements.
Data
minus 1 File (362Kb)
Code
There is no code review at this time.

Reviews
minus Case study of applicationof open source tools. by Luis Ibanez on 09-18-2005 for revision #1
starstarstarstarstar expertise: 5 sensitivity: 4
yellow
Summary:
This paper describes a methodology for registering the successive microscopy slices of a mouse placenta using the author's modified version of the ITK registration framework.

Hypothesis:
The authors suggest that the characterization of placenta morphology can be used for correlating genetic changes with phenotypic variation. The characterization of morphology is intended to be done after reconstructing a 3D model of the placenta via registration of a full microscopy sectioning of a mouse placenta.

Evidence:
The authors provide multiple examples of the images at the various stages of the registration process, and describe many of the details of their registration approach. Unfortunately, the authors do not provide the source code used for their study, nor the images that could be used for replicating their work.

Open Science:
The paper do not satisfy the requirements of Open Science, since it doesnt enable the reader to repeat and replicate the work reported by the authors. The lack of source code and images makes impossible for a reader to replicate the work of the authors. However the paper provide useful advice regarding that image registration methodology that was used by the authors for performing the alignment of the microscopy images.

Reproducibility:
The work cannot be reproduced due to the lack of source code and images.

Use of Open Source Software:
The authors made extensive use of the Insight Toolkit, and modified some of the components in the Registration Framework, in order to suite the particular characteristics of their image registration problem. Unfortunately the source code of such modification is not being shared, so authors that are interested in using a similar approach will have to re-implement the source code for that methodology.

Open Source Contributions:
Source code was not provided

Code Quality:
Source code was not provided

Applicability to other problems:
The authors introduced the interesting concept of a two level optimization scheme where the parameters of the optimizer are set differently at the beggining and at the end of the registration process. This concept seems to be applicable to many if not all of the typical image regsitration problems. It would be interesting to implement this concept into auxiliary classes and to make them available to the community as part of the Insight Toolkit.

Suggestions for future work:
Since the final goal of the authors is to characterize the morphology of the surface of the placenta, it may be interesting to use registration methods that will focus on aligning the boundaries of the placenta. Among the possible options, the authors may want to consider the use of Image metrics that operate on the gradient magnitude of the images, or on the result of edge-detection algorithms such as Canny. The use of PointSet to image registration may also be appropriate for focusing the registration process on aliging the surface of the placenta.

Requests for additional information from authors:
It will be quite useful for readers if the authors share their source code and their image data.

Additional Comments:
The description of the authors reveals that they have dedicated a significant amount of time to this problem. It is unfortunate that readers involved in similar fields would have to reimplement the work reported by the authors due to the lack of source code sharing.

Comment by Ohnatowgabr Ken: yellow


Comment by Ohnatowgabr Ken: yellow


Comment by Ohnatowgabr Ken: yellow


Comment by Ohnatowgabr Ken: yellow


Comment by Ohnatowgabr Ken: yellow


Comment by Ohnatowgabr Ken: yellow


Comment by Tbootsbuy Vigorda: yellow


Comment by Tbootsbuy Vigorda: yellow


Comment by Tbootsbuy Vigorda: yellow

minus Need more .. It doesn't matter if it is validation or code or data or something .. it just needs more by David Holmes on 09-15-2005 for revision #1
starstarstarstarstar expertise: 3 sensitivity: 4
yellow
Summary:
The authors have a large volume serial histology registration problem to address. Like many others, they preprocess the data and register it. The purpose of this paper is not to validate the technique (because there is not enough data or gold standards), but to describe the workflow

Hypothesis:
Implicit hypothesis that there is a workflow (which happens to use ITK) that can process and register these histologic sections

Evidence:
The authors describe the method; however, there is no code and data so there is no real evidence.

Open Science:
The authors state that they use ITK for their processing (they don't state what they use for visualization). There is no code for the two algorithms (processing and registration); however, the registration approach is the basic approach described in all of the ITK documentation. There is not much to contribute back to the open science community because it appears to be a straightforward application of the API. Possibly if the tissue detection algorithm was new, it could be plugged back into the API.

Reproducibility:
Sorry, there is no code. Equally important, there is no data. You would need both for reproducibility. Also, specific parameters for processing would have to be included as well.

Use of Open Source Software:
They used ITK

Open Source Contributions:
Nope

Code Quality:
N/A

Applicability to other problems:
Well, the area of research is important because histologic registration is really needed.

Suggestions for future work:
More validation

Requests for additional information from authors:
I am pleased to see ITK being used for this type of application. It is an important area which the authors are contributing.

I have several comments for the authors' consideration:

(1) Your use of figure is not particularly good. Figure 1 combines two completely different thoughts into one image. Several of the figures (including the left side of Figure 3) are not specificaly referenced in the paper. They should be so a reader can go back and forth.

(2) It appears that some of the pre-processing algorithm has been previously published. I appreciate the fact that your reference that while still giving some of the details. The rest probably needs validation or comparision to other methods. Also, I notice that Figure 2 suggests that you downsample the data. That is not mentioned in the text, but is really important because it has implications for the results.

(3) I like the MI approach that ITK uses. I think that multiple start points is also a good ideal; however, I wouldn't call this a novel idea. It is common to have more than one registration performed with different initial conditions or perturbations.

(4) For this journal (although not all journals), it is really important to have the code and data for independant validation and testing.

Additional Comments:
[This is a free-form field]
Add a new review
Quick Comments


Resources
backyellow
Download All

Statistics more
backyellow
Global rating: starstarstarstarstar
Review rating: starstarstarstarstar [review]
Code rating:
Paper Quality: plus minus

Information more
backyellow
Keywords: Registration, Optimization strategy, Light Microscopy, Mutual Information
Export citation:

Share
backyellow
Share

Linked Publications more
backyellow
An OpenCL implementation of the Gaussian pyramid and the resampler An OpenCL implementation of the Gaussian pyramid and the resampler
by Shamonin D., Staring M.
Confidence Connected Segmentation of the Lateral Ventricles Confidence Connected Segmentation of the Lateral Ventricles
by Radin P.

View license
Loading license...

Send a message to the author
main_flat
ISSN 2327-770X
Powered by Midas