Automatic registration of serial mammary gland sections Page: 4 of 4
This article is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to UNT Digital Library by the UNT Libraries Government Documents Department.
Extracted Text
The following text was automatically extracted from the image on this page using optical character recognition software:
4 of 4
Running the application on a PC Pentium IV (2.66 GHz,
with 512 Megabytes of RAM memory) under Linux, the
standard method of automatic registration on two typical
sections (around 35 Megabytes each) takes between 4 and 5
minutes using 3 different levels of resolution. The shape
registration is faster than the standard method taking
approximately 2 minutes for the same pair of sections.
Usually two pyramid levels are sufficient to obtain a proper
registration. For more complex images, three or more levels
are recommended.
One way of numerically representing the accuracy of the
registration result consists of normalizing the final value of
the matching function. Let m(d, x, y) be the final matching
value of the images being registered (where d, x and y
represent the final rotation angle and translation in x- and y-
axis), and let w, be the number of white pixels in the
thresholded contour image, then the normalized value is
given by:m(d, x, y)= m(d,x,y)
wc x 255(2)
This number, always between 0 and 1, is a pseudo-
percentage of image overlap and allows quantifying the
evolution of the error in every level of the algorithm, and
also comparing both methods.
TABLE I
SHAPE AND STANDARD AUTOMATIC REGISTRATION COMPARISON
Automatic Normalized Matching Function Value
Registration
Method Level 1 Level 2 Level 3Standard 0.92494629 0.94978216
Fig. 1
Shape 0.94635782 0.98543804
Standard 0.96788139 0.98384227
Fig. 2
Shape 0.97498177 0.996523540.949903
0.987002
0.984542
0.997498Table I shows the results obtained with both registration
methods on the sections used for fig. 1 and 2. We can
observe that shape registration provides more accurate
results than the standard method, and that the differences
between level 2 and 3 values are usually quite small.
In some of the mice mammary gland cases, different
types of staining were alternatively applied to the
consecutive sections, increasing the complexity of the
problem due to the different gradients obtained with every
staining. An easy way to solve this inconvenience would
have been to separately and automatically register the
sections with same staining, then manually register two
different stained sections, and last to propagate the
correction to the rest of sections. However, a little
adjustment in the blur filter (reduction in the level of
blurring) allowed us to generalize our automatic method to
all kind of stained sections.IV. DISCUSSION
As stated at the beginning of the paper, consecutive
sections could present non-rigid deformations due to human
interaction in the sectioning and tissue processing. In those
cases, an affine transformation would be insufficient for a
perfect alignment. Local corrections of the rigid registration
would then provide a more accurate approximation to the
problem. Regardless, the rigid registration provides a correct
result in most of our cases, and can be used as the first step
towards a fully non-linear elastic registration in the cases
where linear registration does not provide an accurate
registration.
The systematic search of the functional maximum used
in our method, even if it was improved forcing a 50% image
overlap between images and adjusting the number of
rotations and translations to the image size and pyramid
level, could be replace by an optimization method, thus
approximating the maximum faster and more efficiently.
V. CONCLUSION
We described a fully automatic algorithm for the
registration of microscopy images of consecutive tissue
sections, and a variant of the same method based on the
previous segmentation of the images. The system is based
on a multiresolution and pyramidal approach that calculates
the optimum rigid transformation between the sections. We
have described the algorithm and show results using
Hematoxylin and Eosin (H&E) stained pairs of sections of
mouse mammary glands.
ACKNOWLEDGMENT
I. Arganda-Carreras thanks ImageJ open source project
for provided code (http://rsb.info.nih.gov/ij/).
REFERENCES
[1] Fernandez-Gonzalez R., Jones A., Garcia-Rodriguez E., Chen
P.Y., Idica A. , Barcellos-Hoff M.H., Ortiz-de-Solorzano C. "A
System for Combined Three-Dimensional Morphological and
Molecular Analysis of Thick Tissue Specimens", Microscope
Research and Technique 59(6): 522-530, 2002.
[2] Dean, P., Mascio, L., Ow, D., Sudar, D., Millikin, J. Proposed
Standard for Image Cytometry Data Files. Cytometry 11:561-
569; 1990.
[3] R. Hult, 3-D reconstruction of insect ganglia, Thesis work.
Examensarbete Nr 22, Centre for Image Analysis, Uppsala,
1995.
[4] Fernandez-Gonzalez R., T. Deschamps, Idica A., Malladi R.,
Ortiz de Solorzano C. Automatic segmentation of histological
structures in mammary gland tissue sections. Journal of
Biomedical Optics (accepted, scheduled for publication in May
2004).
Search Inside
This article can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Article.
Arganda-Carreras, Ignacio; Fernandez-Gonzalez, Rodrigo & Ortiz-de-Solorzano, Carlos. Automatic registration of serial mammary gland sections, article, April 13, 2004; Berkeley, California. (https://digital.library.unt.edu/ark:/67531/metadc781669/m1/4/: accessed April 24, 2024), University of North Texas Libraries, UNT Digital Library, https://digital.library.unt.edu; crediting UNT Libraries Government Documents Department.