Related Experiment Video
Updated: Jul 15, 2025
09:31
Automated Quantification of Hematopoietic Cell – Stromal Cell Interactions in Histological Images of Undecalcified Bone
Published on: April 8, 2015
11.6K
Analysis of cellularity in H&E-stained rat bone marrow tissue via deep learning
Smadar Shiffman1, Edgar A Rios Piedra1, Adeyemi O Adedeji1
1Genentech Research and Early Development (gRED), Department of Safety Assessment, Genentech Inc., South San Francisco, USA.
Journal of Pathology Informatics
|September 25, 2023
Summary
This study introduces an automated deep learning method for analyzing rat bone marrow cellularity in preclinical safety assessments. The developed pipeline accurately identifies and quantifies key cell types, aiding hematotoxicity evaluation and drug development.
Area of Science:
- * Computational pathology
- * Preclinical toxicology
- * Hematology
Background:
- * Evaluating cellularity in rat bone marrow is crucial for preclinical safety assessment.
- * Manual analysis of whole slide images is time-consuming and subject to variability.
- * Deep learning offers potential for automating cell segmentation and quantification.
Purpose of the Study:
- * To develop and validate an automated deep learning-based method for assessing cellularity in rat bone marrow.
- * To segment and quantify megakaryocytes (MKCs) and small hematopoietic cells (SHCs).
- * To integrate these models into a pipeline for routine preclinical safety assessment.
Main Methods:
- * Trained a shallow CNN for marrow segmentation.
- * Utilized two Mask R-CNN models for MKC and SHC segmentation.
- * Employed a SegNet model for red blood cell segmentation.
- * Developed a pipeline to identify and count MKCs and SHCs in whole slide images.
- * Compared automated counts and segmentations against pathologist consensus and other deep learning tools (Cellpose, Stardist).
Main Results:
- * The method demonstrated comparable performance to pathologist consensus for MKC segmentation and counting.
- * SHC segmentation and counting showed close median scores with partial overlap in intra-pathologist variation.
- * The automated method provided more accurate SHC counts than Cellpose and Stardist, with a narrower agreement range.
- * The pipeline's performance supports its integration into routine hematotoxicity assessment.
Conclusions:
- * The automated deep learning pipeline effectively evaluates cellularity in rat bone marrow for preclinical safety.
- * This tool can expedite hematotoxicity assessment, aiding drug development.
- * The method enables potential meta-analysis of bone marrow images and generation of new biological insights.
