Abstract Background Flow cytometry plays a critical role in diagnosing hematologic disorders, yet manual analysis is time-consuming and prone to variability. Significant advancements in artificial intelligence (AI) within healthcare have occurred in recent years. Previously, we developed and evaluated an AI-assisted automated system for diagnosing acute leukemia. This study aims to assess the effectiveness of the AI system in diagnosing plasma cell neoplasms. Methods A total of 76 bone marrow samples were analyzed using a panel of antibodies (CD38, CD56, CD138, Kappa, and Lambda) to determine plasma cell percentages and immunophenotype. All samples were used for initial diagnosis of plasma cell neoplasm or therapeutic monitoring. Manual analysis using Kaluza™ software served as the gold standard. The same flow cytometry data was processed by the AI software (DeepFlow™) for comparison. Pearson correlation coefficients were used to compare AI and manual analysis for diagnosis and plasma cell count. Results Among the 76 cases, manual analysis identified 57 positive cases for monoclonal plasma cells (36 cases of CD38+/CD138+/Kappa+ and 21 cases of CD38+/CD138+/Lambda+), with 89% showing aberrant CD56 expression. AI identified 49 monoclonal cases, missing 8 cases with low plasma cell levels (<0.07%) or brighter fluorescence. AI accurately matched 97% of immunophenotypes with manual analysis. Pearson correlation coefficients demonstrated excellent correlation (r=1.0) between manual and AI plasma cell percentages. Furthermore, AI has also enhanced laboratory efficiency, reducing analysis time per case from 10 minutes with manual methods to just 3 minutes with AI analysis. Conclusions This study shows promise for AI-assisted flow software in diagnosing plasma cell neoplasms. AI has enhanced operational efficiency by reducing the manual processing and analysis time. Further training is needed to improve the identification of variations in fluorescence intensity and low plasma cell levels.
Read full abstract