AUTHORS
Wen Yan, MS, Yating Li, MD, PhD2, Yueyi Mu, MB3, Yonghui Xia, MS4, Lu-Gui Qiu, PhD5, Gang An, MD6, Guoqing Zhu, MS7
AFFILIATIONS
1 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
2 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
3 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
4 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
5 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
6 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
7 State Key Laboratory of Experimental Hematology, National Clinical Research Center for Blood Diseases, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, China
ABSTRACT
Background Accurate assessment of circulating plasma cells in peripheral blood is clinically relevant across plasma cell disorders, including multiple myeloma and plasma cell leukemia. Conventional manual morphology assessment relies on evaluation of a limited number of leukocytes and may be affected by sampling variability, particularly when plasma cells are present at low abundance. We evaluated an artificial intelligence (AI)-based whole-slide analysis approach for plasma cell quantification by increasing cellular assessment depth and assessed its agreement with expert morphology review and flow cytometry.
Methods We retrospectively analyzed 250 peripheral blood smears from consecutive patients with plasma cell disorders undergoing clinical evaluation at our center between March 12, 2025, and January 20, 2026. The cohort included multiple myeloma, plasma cell leukemia, monoclonal gammopathy of undetermined significance, smoldering myeloma, lymphoplasmacytic lymphoma/Waldenström macroglobulinemia, and other plasma cell neoplasms.
Each smear was independently reviewed by three certified hematology laboratory technologists using blinded manual differential counting of 200 leukocytes per case. Whole-slide images were generated through automated scanning of evaluable smear regions and analyzed using an AI-based digital morphology system. AI-derived plasma cell percentages were compared with manual morphology assessment. In a subgroup of 135 patients with paired peripheral blood flow cytometry, AI-derived plasma cell burden was compared with total plasma cell burden measured by flow cytometry.
Results Whole-slide AI-based digital morphology analyzed a median of 14,682 leukocytes per case (interquartile range [IQR], 10,884–19,910), representing approximately 70-fold greater cellular assessment depth than conventional manual leukocyte review.
Interobserver agreement among three certified hematology laboratory technologists was excellent (ICC=0.97), supporting manual morphology as a robust reference standard. AI-derived plasma cell percentages demonstrated strong agreement with manual morphology assessment (ICC=0.91; 95% CI, 0.89–0.93). Bland-Altman analysis showed minimal systematic bias between AI and manual assessment (mean difference, −0.001%; limits of agreement, −3.94% to 3.94%).
Among 114 cases with no plasma cells identified by manual morphology assessment, AI detected measurable plasma cells in 68 cases (59.6%), with a median AI-derived plasma cell proportion of 0.02% (range, 0.01–0.41%), demonstrating enhanced detection capability at very low plasma cell levels.
In 135 cases with paired flow cytometry, AI-derived plasma cell burden showed excellent agreement with flow cytometric quantification (Spearman ρ=0.91; ICC=0.99). Passing-Bablok regression demonstrated consistent proportional agreement between AI and flow cytometry (slope, 0.62; 95% CI, 0.59–0.70). AI accurately discriminated flow cytometry-defined plasma cell positivity (AUC=0.90) and achieved excellent performance for clinically relevant thresholds of ≥0.1% and ≥1% plasma cells (AUC=0.986 and 0.990, respectively). Among 46 AI-positive/manual-negative cases with paired flow cytometry, 40 (87.0%) demonstrated detectable circulating plasma cells by flow cytometry, supporting enhanced analytical sensitivity of AI-based assessment.
Conclusion AI-based whole-slide analysis provides reliable plasma cell quantification in peripheral blood specimens from patients with plasma cell disorders, demonstrating strong agreement with expert morphology assessment and flow cytometry. By substantially increasing cellular assessment depth compared with conventional manual counting, AI analysis enables detection of low-abundance circulating plasma cell populations that may be missed by routine morphology review. This approach may serve as a complementary tool to enhance peripheral blood evaluation in plasma cell disorders.
DISCLOSURES
Wen Yan, MS: No relevant conflicts of interest to declare.