Background and Objectives: The use of conventional microscopy still forms the basis for the morphologic evaluation of erythrocytes despite widespread use of automated tests in the hematology laboratory. This requires a considerable length of time and expertise, and have the potential of becoming a source of errors and delay in reporting. Advances in image processing and machine learning in recent years have shown acceptable performance characteristics and have promising applications in the diagnostic laboratory. Use of these newly-developed technologies can address the stated problems and provide an alternative approach in the microscopic analysis of erythrocytes. Methodology: This prospective validation study compared digital image analysis using a machine-learning based image recognition algorithm with conventional microscopy performed by a trained microscopist, which served as the reference standard. Random deidentified anticoagulated peripheral blood samples submitted to the hematology laboratory were assessed. Results: A total of 956 erythrocytes were evaluated after image processing using support vector machine and routine microscopy as classifiers of erythrocytes into three categories: size, central pallor, and shape. The tested software was able to achieve a strong level of agreement compared to conventional microscopy, having kappa values ranging from 0.81 to 0.86. Accuracy for size, central pallor and shape were 89.88%, 93.72% and 87.89%, respectively. Conclusion: The validated image recognition software is an acceptable diagnostic test in determining erythrocyte morphology in peripheral blood smears. Its integration can potentially minimize hands-on time and improve the diagnostic laboratory workflow. Registration Philippine Health Research Registry (PHRR) ID: PHRR191211-002348; University of the Philippines Manila Research Ethics Board (UPMREB): 2019-356-01.
Microscopy