Developing a Digital Analysis Algorithm To Count Intraepithelial Eosinophils on H&E Stained Sections for Diagnosis of Eosinophilic Esophagitis.
Vamsi Parimi, Anna Zago, Ximing J Yang, Zongming E Chen. Northwestern University, Chicago, IL
Background: Eosinophilic Esophagitis (EoE) is a distinct clinicopathologic entity with classic symptoms and endoscopic findings. Histologically, it is characterized by increased infiltration of eosinophils in the esophageal squamous mucosa. Quantification of intraepithelial eosinophil is a major diagnostic criterion. Manual enumeration has been shown not only tedious but also inconsistent among observers. We attempt to develop an automated counting algorithm using Aperio Genie program.
Design: Five H&E stained slides of esophageal biopsies from EoE patients were selected and scanned at 20X objective with Aperio Scanscope automated digital scanner. An automated counting algorithm was developed by training Aperio Genie program to recognize characteristic features of eosinophil cytoplasm and nuclear size. To validate the accuracy, the algorithm was applied to selected fields of the digitally scanned slides. In each field, computer generated positive events were visually located and compared to the corresponding H&E image. False positive and negative events were identified and a concordance rate was calculated by dividing the number of computer generated correct events to the number of visually identified eosinophils.
Results: After testing multiple settings, we developed an optimal algorithm incorporates both features of eosinophilic granular cytoplasm and nuclear size exclusion criteria as classifiers. This algorithm was applied to 21 selected areas covering all five scanned H&E sections. The concordance rate ranged from 57% to 100%, with a mean of 84+14 %. False positivity was primarily due to red blood cells contamination and eosin staining artifact at the edge of tissue fragments. Overall false negativity was 9.7% and was mainly due to less well revealed eosinophil cytoplasm. By eliminating these factors, the concordance rate reached to 100% in all 5 additional tested fields.
Conclusions: We have demonstrated the feasibility for developing an algorithm for automated enumeration of eosinophil on H&E slides with great accuracy. This is potentially useful for standardization of EoE diagnosis and research.
Monday, February 28, 2011 1:00 PM
Poster Session II # 279, Monday Afternoon