[1354] Automated Image Analysis of Cytoplasmic Immunohistochemical Stains: A Technique To Overcome Thresholding Challenges

D Gui, G Gerney, N Doan, G Cortina, S Ohsie, J Said, S Dry. UCLA, Los Angeles

Background: Quantitative immunohistochemical analysis may be more efficient and reproducible than manual interpretation. Unfortunately, image segmentation/thresholding does not work as well for cytoplasmic staining. The brown diaminobenzidene (DAB) chromogen contains red, blue and yellow hues; hematoxylin, the usual counterstain, is blue. When hematoxylin-counterstained images are thresholded for DAB, resulting masks include nonspecific areas, due to the blue component that is part of the DAB brown. No-counterstained images may improve accuracy of quantitative IHC for DAB labeled cytoplasmic antigens.
Design: Three small bowel samples (one section per sample) were analyzed. Initially, the slides were stained for serotonin (DAB) with no-counterstain (NC) and scanned; then they were counterstained with hematoxylin (HC) and scanned again. Scanning was performed at 20x using an Aperio XT wholeslide scanner. The area of cytoplasmic stain and total tissue stained was analyzed using Metamorph software (v 7.0.r4). Five 20X fields per case (15 areas total) were analyzed. The identical areas were analyzed with DAB-NC and DAB-HC.
Results: The automated total measured area was about twice the size using HC compared to NC (17833.14 mm² for HC vs. 9425.99 mm² for NC). The number of positive cells was identical by manual counts in both groups (n=205). The difference in area counted by the computerized analysis was statistically significant (t-test, p value = 0.002). Since the exact same areas were analyzed with and without HC, the difference in automated measured area can be attributed to thresholding error.
Conclusions: Use of no-counterstained (NC) images improves accuracy of image analysis for cytoplasmic immunostains that use the DAB chromogen. These findings are important for researchers performing quantitative immunohistochemistry.
Category: Informatics

Wednesday, March 11, 2009 9:30 AM

Poster Session V # 218, Wednesday Morning