Abstract: Background – Hematocrit (HCT) and red blood cell (RBC) count are crucial indices that assist in the diagnosis and monitoring of hematological disorders such as anemia and polycythemia. Rapid and accurate assessment of these parameters is essential. Hypothesis/Objectives – The aim of this study is to estimate HCT and RBC count from microscopic blood smear images. It is hypothesized that there is a significant agreement between these indices derived from the images and from the flow cytometry-based hematology analyzers. Animals – Microscopic blood smear images from 986 dogs and 1522 cats. Methods – Correlational study. Features of a blood smear are used to train a neural network to predict hematocrit and RBC count. The validity and accuracy of the proposed method are assessed by evaluating its correlation with the results from the flow cytometry-based hematology analyzer (ADVIA® 2120 Hematology System, Siemens Healthcare Diagnostics, Tarrytown, NY, USA). Results – There is a significant correlation between image- and ADVIA-derived hematocrit estimate in dogs (p<.0001, r = 0.71) and cats (p<.0001, r = 0.73). Similarly, a strong positive correlation is observed between image- and ADVIA-derived RBC count estimate in dogs (p<.0001, r = 0.71) and cats (p<.0001, r = 0.66). Conclusions and clinical importance – The image-based indices are in strong agreement with those from flow cytometry analyzers. Image-based analysis of hematological indices can help clinicians evaluate for RBC disorders rapidly and accurately even in the absence of conventional hematology analyzers.
