Evaluation of Artificial Intelligence Driven Digital Morphology for Estimation of Platelet Count: A Cross-sectional Study

Introduction: Nowadays different technologies are employed in the field of automated haematology analyser that provides rapid and reliable estimations and is regularly used for determining the platelet counts. In some cases, it gives erroneous results, especially with Red Blood Cells (RBC)-platelet interference and giant platelets, which need to be verified by manual methods. The traditionally used manual method microscopic estimation of platelet on smear is labour-intensive, produces variable results and is subject to observer bias. Aim: The present study aimed to compare platelet count estimation by SigTuple-AI100 ShonitTM (AI100)- a digital morphology analyser with manual platelet counts and automated haematology analyser (HORIBA Yumizen H2500). Materials and Methods: The present cross-sectional comparative study done in Department of Haematology, Suburban Diagnostics Referral laboratory, Mumbai, India for a period of eight months from June 2023 to February 2024. One hundred Ethylenediaminetetraacetic Acid (EDTA) whole blood samples were analysed for platelet count by an automated haematology analyser (HORIBA Yumizen H2500), manual microscopic method, and digital morphology platform (SigTupleAI100). Results were analysed using IBM Statistical Package for Social Sciences (SPSS) Statistical Software version 26. Estimated platelet counts of AI100 were compared with manual platelet count and Yumizen H2500, using Pearson’s correlation coefficient and Bland-Altman plot analysis. Manual platelet count was used as a reference method. Results: Platelet counts from the AI100 system showed an R2 of 0.91 when compared to manual platelet estimates and an R2 of 0.92 when compared to the automated haematology analyser results. Conversion factor was derived and validated on 100 consecutive thrombocytopenic samples. Coefficient of Variance (CV%) of AI100 was 4.95. Conclusion: The study suggested that the platelet count obtained via the AI100 compared well with both automated haematology analyser and manual method. Though similar platforms are available worldwide, the cost of AI100 is cheaper than other platforms. There is also ease of operations and it being cloud based allows skilled Pathologists and lab technologists to report remotely