Automatic Bone Marrow Cellularity Estimation in H&E Stained Whole Slide Images

Area of Science:

• Digital pathology
• Hematopathology
• Computational imaging

Background:

• Bone marrow cellularity estimation is crucial for hematopathology diagnostics.
• Manual assessment of cellularity from whole slide images (WSI) is subjective and prone to variability.
• Computer-assisted methods can potentially improve objectivity and reproducibility.

Purpose of the Study:

• To develop a fully automated algorithm for estimating bone marrow cellularity in H&E stained WSIs.
• To utilize bone marrow segmentation for accurate cellularity assessment.
• To reduce inter- and intraobserver variability in bone marrow cellularity estimation.

Main Methods:

• Developed an algorithm involving biopsy segmentation, tissue classification, and bone marrow segmentation.
• Segmented red and yellow bone marrow (YBM) regions within H&E WSIs.
• Validated segmentation using DICE coefficients and compared automated cellularity estimates with manual assessments using Bland-Altman plots and ICC.

Main Results:

• Achieved average DICE scores of 0.901 for red marrow and 0.920 for YBM.
• Reported mean cellularity estimate differences of -0.552 and -7.816 compared to two manual estimates.
• Obtained a high intraclass correlation coefficient (ICC) of 0.980 between automated and manual estimates.

Conclusions:

• The automated algorithm demonstrates a good agreement with expert manual bone marrow cellularity estimates.
• This computational approach offers a promising tool for objective and reproducible hematopathology analysis.
• Fully automated bone marrow cellularity estimation is feasible and can aid diagnostic accuracy.