The initial goal of this project was to adjust a number of the nonsymmetric color cryosection images of the National Library of Medicine's Visible Human Data Set. To do this we rotated the images by a small angle, splicing appropriate image sections together. We then moved on to investigate feature
extraction methods based on color in order to isolate a specific
tissue type from the cryosection images of a male cadaver. This work is part of the Visible Human Project, which will enhance the undergraduate study of anatomy. Students will be able to access and manipulate the cadaver data from the type of workstation found on the average college campus as opposed to
access being limited to students at institutions which possess the high
powered workstations for which most manipulation of these large data
sets has previously been done. Simple, rapid segmentation techniques
based on color will aid in viewing individual organs and tissues on such
Two methods of tissue extraction were compared. The first
was a histogram intersection method. The image was
traversed and histograms of extracted pixel blocks were compared.
The program allowed us to manipulate the number of bins in the
histograms, the comparison threshold, the size of the blocks, number
of library images used, and the histogram's color space (OPP vs.
HSV). The second method segmented images through a Zero Crossing
edge detector and a noise reduction program. Thus, we allowed for
non-square images to be compared with the library images. In both
cases the sections that exceeded the given threshold were then
written to an outfile image.
For the tissue retrieval we ran our tests searching for fat
and muscle on an abdominal image which also contained liver, kidney, and
bone, as well as several other tissues. Our nonsegmented method, in
its attempt to isolate muscle, was unable to keep parts of the liver and
kidneys from being interpreted as muscle tissue. Setting the threshold
to a higher value eliminated some of the detected muscle tissue along
with the unwanted liver and kidney tissue. Using the two different
color spaces gave useful results, but was still plagued by the
detection of incorrect tissue. Using the Zero Crossing edge detector for
preliminary image segmentation succeeded somewhat in separating the
segmentation process from the tissue identification but, due to
incomplete segmentation, resulted in some ambiguities in tissue
This work has shown that standard image processing techniques
based on color can be extended to isolate specific tissues from
cryosection images. The algorithms used here were implemented on readily
available workstations and shown to execute in minimal time.
In conclusion, since much of human tissue has similar color
characteristics, color based differentiating will always run into
problems of misdetecting other tissue. However, we have found that
although color may not be a strong enough characteristic to perfectly
extract tissue from cryosection images, it is a strong key for image
segmentation and tissue identification. Incorporating color segmentation
techniques into those used for texture, location, or other attributes
could minimize the problems inherent with each individual method.
tech report (4.5 MB)
A senior at Augustana College, Charles Haben is majoring in Physics, Mathematics, Computer Science, and Secondary Education. He has served as Assistant Residence Director and captain of the Track and Field team. He recently presented a project on utilizing local campus resources at the combined Indiana and Illinois Physics Teachers Conference.
A junior at the College of St. Catherine, Rose Hennessy is majoring in Mathematics and International Business and Economics, and minoring in Computer Science. She has served as a Mathematics tutor, treasurer of the business and economics clubs, and administrative coordinator of the campus programming board. Her work on CT Data Manipulation for Cryosection Enhancement was presented at the 1997 Argonne National Laboratory Symposium for Undergraduates in Science, Engineering and Mathematics.