Color deconvolution for the analysis of tissue microarrays

Abstract

OBJECTIVE: To analyze tissue microarrays (TMAs) using color deconvolution, a method for separating component dyes in digital images, and compare the results to observer scoring. STUDY DESIGN: TMAs were constructed from tissues from 100 adult autopsies and immunohistochemically stained for connective tissue growth factor. A region of interest (ROI) was created for each core image using 3 binary masks-tissue area, inclusion area and exclusion area. The diaminobenzidine (DAB) and hematoxylin signals were deconvolved, and the DAB signal was measured for each ROI. The dorsalis pedis core images were also scored manually. RESULTS: Seventeen TMAs were annotated, requiring 532 minutes. Of the 1,683 cores, 296 (18%) were excluded because they were not suitable for evaluation. A single TMA required a mean of 31.3 minutes to evaluate; to annotate a single core took a mean of 19 seconds. For the dorsalis pedis, observer score and median DAB intensity correlated strongly (Kendall’s = 0.71). CONCLUSION: Analysis of TMAs by color deconvolution is efficient and highly correlates to observer scoring.

Publication
Anal Quant Cytol Histol
Toby C. Cornish
Toby C. Cornish
Professor of Pathology and Data Science Institute

Clinical informaticist, gastrointestinal pathologist, and researcher.

Related