Deep Learning-Based Open Source Toolkit for Eosinophil Detection in Pediatric Eosinophilic Esophagitis.

Area of Science:

• Medical Imaging
• Computational Pathology
• Immunology

Background:

• Eosinophilic Esophagitis (EoE) is a chronic immune-mediated esophageal disease.
• Manual histological analysis of eosinophil-dominant inflammation in whole slide images (WSIs) is labor-intensive and prone to inaccuracies.
• Accurate detection of eosinophils is critical for EoE diagnosis.

Purpose of the Study:

• To develop an open-source, automated toolkit for eosinophil detection in WSIs for EoE diagnosis.
• To leverage deep learning and ensemble strategies for enhanced detection accuracy and reliability.
• To provide a user-friendly, command-line tool for efficient analysis.

Main Methods:

• Development of Open-EoE, an open-source toolkit utilizing Docker for end-to-end WSI analysis.
• Implementation of three state-of-the-art deep learning object detection models.
• Application of an ensemble learning strategy to optimize model performance and reliability.

Main Results:

• The Open-EoE toolkit efficiently detected eosinophils (Eos) across a test set of 289 WSIs.
• Achieved 91% accuracy in detecting eosinophils at the diagnostic threshold of ≥ 15 Eos per high power field (HPF).
• Demonstrated consistent performance compared to pathologist evaluations.

Conclusions:

• Open-EoE offers a promising, automated approach for eosinophil detection in EoE diagnosis.
• Machine learning integration can significantly improve the efficiency and accuracy of EoE diagnostics.
• The open-source toolkit facilitates wider adoption and further research in computational pathology for EoE.