Validation of a QuPath-Based Artificial Intelligence Workflow for Plasma Cell Quantification on CD138
Mohammed N Meera1, Ankur Ahuja1, Venkatesan Somasundaram1
1Department of Pathology, Armed Forces Medical College, Pune, India.
Insights
AI-assisted QuPath analysis of CD138 immunohistochemistry in bone marrow biopsies shows high accuracy for plasma cell quantification. This digital pathology approach overcomes manual estimation variability, improving diagnosis and risk stratification for plasma cell neoplasms.
Area of Science:
- Hematopathology
- Digital Pathology
- Artificial Intelligence in Medicine
Background:
- Accurate plasma cell quantification in bone marrow biopsies is crucial for diagnosing and staging plasma cell neoplasms.
- Manual estimation of plasma cell percentage via CD138 immunohistochemistry is prone to inter-observer variability.
- An AI-assisted whole-section workflow offers a potential solution to enhance objectivity and reproducibility.
Purpose of the Study:
- To validate an AI-assisted whole-section analysis workflow using QuPath v0.6 for plasma cell quantification.
- To compare the AI-assisted method against conventional manual assessment in an Indian hematopathology setting.
- To evaluate the diagnostic performance of the AI-assisted method at International Myeloma Working Group (IMWG) thresholds.
Main Methods:
- Fifty CD138-immunostained bone marrow trephine biopsies were analyzed.
- Methods included manual overview estimation, AI-assisted whole-slide image analysis with QuPath v0.6, and systematic cell-by-cell manual counting (reference standard).
- Concordance was assessed using Spearman ρ, ICC, Lin's CCC, Bland-Altman analysis, Passing-Bablok regression, and weighted Cohen's κ.
Main Results:
- The AI-assisted QuPath method demonstrated excellent concordance with the manual reference standard (ICC=0.995, Lin's ρc=0.995, Spearman ρ=0.989).
- Bland-Altman analysis showed minimal bias (+2.4 percentage points) with narrow limits of agreement.
- Diagnostic performance at IMWG thresholds was excellent, with an AUC of 0.995-1.000 and 100% sensitivity at all thresholds.
Conclusions:
- AI-assisted QuPath-based plasma cell quantification is highly accurate and reproducible.
- The digital pathology workflow significantly reduces inter-observer variability compared to manual estimation.
- This AI tool shows strong diagnostic performance, supporting its clinical utility in managing plasma cell neoplasms.
Introduction:
Plasma cell percentage on CD138 immunohistochemistry of bone marrow trephine biopsies is fundamental to the diagnosis and risk stratification of plasma cell neoplasms, yet manual visual estimation is subject to inter-observer variability. We report a single-center pilot validation of an AI-assisted whole-section workflow using the open-source platform QuPath v0.6 against conventional manual methods in an Indian tertiary hematopathology setting.
Methods:
Fifty consecutive CD138-immunostained trephine biopsies were independently assessed by (A) overview estimation by three hematopathologists, (B) digital whole-section analysis with QuPath v0.6 on whole-slide images acquired with the OptraScan OS-Ultra scanner, and (C) systematic cell-by-cell manual count by the principal investigator (reference standard). Concordance was assessed by Spearman ρ, intraclass correlation coefficient (ICC), Lin's concordance correlation coefficient (CCC), Bland-Altman analysis, Passing-Bablok regression, and weighted Cohen's κ. Diagnostic performance of Method B was evaluated at the International Myeloma Working Group (IMWG) thresholds of 10%, 30%, and 60%.
Results:
Method B versus Method C showed ICC = 0.995 (95% CI, 0.97-1.00), Lin's ρc = 0.995, Spearman ρ = 0.989, Bland-Altman bias +2.4 percentage points (95% limits of agreement -3.0 to +7.8). Passing-Bablok regression gave slope 1.025 (95% CI, 1.00-1.06) and intercept +1.54, indicating absence of proportional bias. Diagnostic performance at IMWG thresholds was excellent (area under the curve (AUC) 0.995-1.000; sensitivity 100% at every threshold). Weighted Cohen's κ for three-grade burden classification was 0.91 with no extreme misclassifications.
Conclusion:
AI-assisted QuPath-based whole-section plasma cell quantification using CD138 immunohistochemistry achieves excellent concordance with manual reference counting and strong diagnostic performance across all clinically decisive IMWG thresholds.
