This page is best viewed in Firefox.

Imaging: Research

Our research and development of biomedical imaging techniques can be broken down into a few fundamental categories.


A count of cell nuclei using Hoechst 33352.
Blue: original image; Green: counted points.

Some cellular events such as proliferation and apoptosis can be characterized by single points on an image. The relevant data relating to this sort of image is most accurately represented by binary identification. The images produced through staining are not binary however, and steps must be taken to reduce the image to a point map which can then be used to gather useful information. Difficulties in doing this relate to varying stain intensities, bleedthrough effects, and the high proximity in which these events can occur.

Once these points have been identified (as shown in the picture to the right) a number of useful analyses can be run relating these points to other images via location or relative number in a given region. Having multiple stains on a single section permits a number of interesting comparisons. We have typically focused on proliferation relative to different hypoxia levels, but this could easily be adapted to a number of similar methods.


Hoechst Mask

A basic step in almost any image analysis technique is the identification of regions through masking. The general idea is to create a region or regions of interest from an image resulting in a binary mask.

To start any analysis, it is necessary to first determine where the tissue is located. This helps to correct for (and identify) background luminescence levels which are useful in correcting the actual image intensity values. Much of our masking is currently based on images stained with Hoechst 33352 which identifies nuclei. We are currently in the process of refining our methods to produce a map that also locates necrotic regions in the tissue. More accurate tissue masks can be created through the addition of a hemotoxalin and eosin image.

Object Identification:

Many structural components in tissue can be visualized through staining procedures. It is often important to identify them in a way similar to masking. Some difficulties arise in getting a clear picture of each structural object as differentiating it from the background can be tricky.

Vessel Identifying
Top Left: Original Image; Top Right: Threshold Identification;
Bottom Left: "Advanced" Technique; Bottom Right: Result of advanced identification process.

The above images show the results of two approaches to object identification. The top two images are before and after depictions of simple thresholding performed on a CD31 stained image. While this does clearly identify vessels, it often fails to encompass the entire vessel. The bottom two images are the results of applying a more rigorous approach that attempts to connect nearby "possibly" related components. The algorithm does a very good job of determining which structures are likely linked and generates an image which more accurately depicts the vascular structure.

Region Mapping:

Our third major imaging approach is a calibrated and quantifiable approach to region mapping. With the use of EF5, we are able to correlate observed intensity values to oxygen pO2 values on a per pixel basis. It is then possible to bin the image into regions defined by differing oxygen levels. Analyses can then be generated based on this new region map, or simply using the calibrated pO2 values in relation to other identified region maps.

Oxygen Map

The image to the left is an example of an oxygen map which uses a binning process to isolate various regions based on their local pO2 values. A Hoechst mask has been used to denote off tissue areas in black. Oxygen rich areas are colored teal and increase in redness and intensity toward yellow and white as oxygen decreases.

Comments? Questions? E-mail