Biomedical Image Processing

Imaging is a powerful and indispensable type of methods in biomedical and medical research. Doctors and researchers rely on PETs/CT, functional MRI, western blots, microscopic images to examine patients and cell functions. However, most of these images are analyzed by human eyes qualitatively and manually, and the finer, quantifiable information of the images can be lost in the qualitative comparison.

For example, we hypothesized that cancer cells can drive the “browning” of adjacent adipose tissue and obtained immunohistochemistry image from a tumor tissue, here “browning” means the UCP1 expression stained in brown, shown in Figure 1. Although one could roughly tell that the random brown regions permeate more than that on the right region of the image, it is difficult to tell how much. This challenges us to reject the null hypothesis confidently. By quantifying the brown pixels using image processing techniques, we could quantitatively show precisely the decreasing trend of UCP-1 expression with respect to the proximity of cancer cells, shown in Figure 2. This image processing method could be automatically applied to the other slices of same set of immunohistochemistry images, and the researchers could gather a summary statistic information from the images. Similar analysis techniques could be applied to wester blots and other types of microscopic images.


Figure 1. Inguinal WAT adjacent to mammary tumor appears to have smaller lipid droplets and increased UCP1 staining intensity compared to ovarian WAT, resembling the beige phenotype. Image was taken from one representative mouse at 10X magnification.

Figure 2. Analysis of the number of brown pixels, proxy for UCP1 staining intensity, demonstrating maximal staining at the tumor-adipose interface and decreasing with distance from the tumor.


Another example that computer-aided image processing techniques could contribute to medicine is to automate the processing of radiological images.