6th ESACP Congress, Heidelberg, April 7-11, 1999

A045
QUANTIFICATION OF IMMUNOHISTOCHEMISTRICALLY STAINED CELL NUCLEI OF BREAST CARCINOMA TISSUE SECTIONS
Choi HK 1, Heo MK 1, Choi HJ 1, Nam SH 2, Seo JW 3

1) Department of Computer Sience and 2) Biomedical Engineering, Inje University, Kim-Hae Kyung-Nam, 3) Department of Pathology, Seoul National University, Chongu-Gu, Seoul, Korea

We analyzed immunohistochemistrically stained cell nuclei for a color image of breast carcinoma tissue sections taken from patients to diagnose objectively the degree of the progression. The purpose of our research is to detect the degree of the distribution of positive objects and negative objects on an image. The digitized images(40x) were obtained 4 µm thick paraffin wax sections and a pixel size about 0.4 µm. The color images showed a difficulty to distinguish between light brown color and light blue color because we transformed from RGB to HLS for visualization. Our analysis is based on two different methods. One is to compute the area proportions of the positive objects. The other is to count cell nuclei of the positive objects. We used a thresholding histogram of three RGB band images and box classifications in a RGB color space to calculate object areas. However, the segmented image appeared with noises and holes in the objects. Small objects of the area less than 60 pixels were removed and filled with holes. We obtained the result 21.2% for the training set. The results of test images obtained 52.2% and 55.6%. The total segmented cell nucleus number divided by the positive cell nucleus number and the results obtained 17% for the training set. The test image obtained 57% and 73%. These results have a good correlation with visual inspection by pathologist.