A-169 Artificial Intelligence based Complete Blood Count

Abstract

Abstract Background The OLO device (Sight diagnostics, Israel) is an artificial intelligence based blood counter. OLO contains a fluorescent microscope and a camera embedded inside it. The pictures are processed by an image processing algorithm based on “machine-vision technology” to identify and quantify cells. The measurement analysis is accompanied by messages (flags) that are used to warn of deviations in the counted parameters, which help the operator refer the sample for microscopic analysis. In this study, we compared the results of 19 complete blood count (CBC) parameters obtained from OLO to the CBC parameters obtained from the ADVIA 2120 device (Siemens, USA). In addition we examined the reliability of OLO alerts (flags) by comparing it to the microscopic analysis findings. Methods 99 blood samples from patients aged >3 months were sampled by the ADVIA and OLO devices. Regression analysis was performed in comparing the results of 19 parameters between the devices. Specificity and sensitivity values ​​of the OLO device alerts were calculated. Results A good correlation was found for CBC parameters between OLO and ADVIA with Pearson correlation (R) of 0.992 (WBC), 0.985 (RBC), 0.991 (HB) and 0.983 (PLT). A fundamental difference in measurement and classification technologies between OLO and ADVIA lead to low correlation measured for monocytes. By comparing OLO alerts against blood smears findings, a total of 14 samples were alerted by OLO but only slight left shift/single pathological cell were found in the smears which is below threshold defined by the manufacturer. These 14 samples were transferred to the true positive category, therefore OLO device alerts showed sensitivity and specificity of 81.8% and 90% (respectively) in identifying pathological cells. Conclusions The comparison findings present a technology based on image processing and artificial intelligence that have a good correlation to the ADVIA reference device. In addition, OLO has the ability to warn of pathological findings even below the threshold defined by the manufacturer. The artificial learning is expected to be improved so that this device can integrate CBC and microscopic differentiation simultaneously, a great advantage to the hematology laboratories workflow.