Code-Free Development and Deployment of Deep Segmentation Models for Digital Pathology.

Area of Science:

• Computational pathology
• Digital pathology
• Machine learning in histopathology

Background:

• Deep learning for histopathological whole slide images (WSIs) can enhance diagnostic accuracy but requires coding skills or commercial software.
• Accessibility to advanced computational pathology tools is limited for researchers without programming expertise.

Purpose of the Study:

• To present a code-free pipeline for developing and deploying deep learning segmentation models for computational pathology.
• To enable pathologists without programming experience to create state-of-the-art segmentation solutions for WSIs.
• To showcase the utility of open-source software in creating generalizable and accessible computational pathology pipelines.

Main Methods:

• Utilized a code-free pipeline with open-source software (QuPath, DeepMIB, FastPathology) for model creation and deployment.
• Developed a dataset of 251 annotated WSIs (140 HE, 111 CD3) for colon biopsy epithelium segmentation using active learning.
• Evaluated model performance on a hold-out test set of 57 WSIs.

Main Results:

• Achieved high mean intersection over union (IoU) scores of 95.5% (HE) and 95.3% (CD3) for epithelium segmentation.
• Demonstrated pathologist-level segmentation accuracy and clinically acceptable runtime performance.
• Validated that pathologists without programming experience can create advanced segmentation solutions using free software.

Conclusions:

• The presented code-free pipeline democratizes deep learning application in computational pathology, making advanced tools accessible.
• Open-source solutions facilitate the creation of generalizable, exportable segmentation models and pipelines, accelerating research.
• The pipeline, including models and data, is openly available to foster further research and development in digital pathology.