Eosinophilic Cell Counting in Pediatric Ulcerative Colitis using Histopathology Images

Abstract:

The PROTECT study identified rectal mucosal eosinophilia to be associated with increased odds of achieving week 4 remission and reduced odds of escalating to anti-TNF therapy by one year. To apply these insights in a clinical setting, a reliable and scalable eosinophil cell counting approach needs to be developed. Relying on pathologists to manually perform a peak region count is restrictive, time consuming, and leads to variation in counting.  In this study we developed a state-of-the-art deep learning cell-classifier tool to automate rectal eosinophilic counting.  To train a generalizable eosinophil cell classifier tool, we had four IBD pediatric pathologists from two pediatric hospitals (HSC & CCHMC) annotate 584 crops from 98 standard-of-care hematoxylin and eosin (H & E) diagnostic treatment naïve rectal mucosal biopsies. The test set also consisted of 54 annotations by two pathologists.  Pixel-wise AUROC for eosinophil cell classification was high on all data sets: 0.96 (95%CI:0.95-0.96), validation 0.95 (95%CI:0.94-0.96) and test set 0.94 (95%CI:0.94-0.95). The inter-rater reliability measured using spearman correlation coefficient among the two pathologists was 0.96 (95%CI0.93-0.97). The correlation between the pathologist and the algorithm slightly less at 0.89 (95%CI:0.82-0.94) and 0.88 (95%CI:0.80-0.94) respectively. Using the post-processing clustering technique, the R-squared between the predicted and actual cell count improved from 0.58 (95%CI:0.42-0.71) to 0.8 (95%CI:0.71-0.86) over the baseline U-Net output.

References:

J. Reigle, O. Lopez-Nunuz, E. Drysdale, D. Abuquteish, X. Liu, J. Putra, L. Erdman, A. M. Griffiths, S. Prasath, I. Siddiqui, J. Dhaliwal. Using deep learning to automate eosinophil counting in pediatric ulcerative colitis histopathological images. medRxiv, April 2024. 

Preliminary version at medRxiv: 10.1101/2024.04.03.24305251

Posters/Abstracts/Presentations:

E. Drysdale, X. Liu, J. Reigle, O. Nunez-Lopez, I. Siddiqui, T. D. Walters, J. S. Hyams, L. A. Denson, S. Prasath, J. Dhaliwal. Employing deep learning approaches to automate eosinophilic cell counting in pediatric UC. Digestive Disease Week (DDW), San Diego, CA, USA, May 2022. Poster presentation.

Back to Histopathology Projects Main Page. Back to Research Page.