Abstract:
One of the main sources of errors when conducting biochemical analysis of blood serum in a clinical diagnostic laboratory is the excessive concentration of hemoglobin (hemolysis) or lipids (lipemia) in the analyzed sample. Therefore, an important step to accurately determine the concentration of the target analyte is to first classify the sample into “suitable” and “unsuitable” classes for analysis. At the same time, to be used in practice, the method of preanalytical classification of samples must be both simple to implement and reliable, from the point of view of high sensitivity and specificity. In this work, we investigated the analytical ability of two approaches – an approach based on diffuse reflectance spectroscopy, characterizing the parameters of diffuse reflection of blood serum in the visible and near-IR range (500–1000 nm), and an approach based on computer vision – in classifying blood serum samples for normal suitable for analysis, and samples with hemolysis and lipemia. Diffuse reflectance spectroscopy has been found to demonstrate high sensitivity and specificity (more than 97%) in the classification of serum samples, but technically this method requires the application of a measuring probe to the sample. At the same time, computer vision methods have made it possible to determine the suitability of a sample for further analysis with lower classification accuracy values, but in more complex conditions, in particular, in the case of a sample moving along a conveyor line in a clinical diagnostic laboratory. The advantage of the studied methods, in addition to the high accuracy of preanalytical classification, is the simplicity of their technical implementation, as well as the ability to characterize samples without additional sampling of blood serum, which indicates their promise as methods for preanalytical analysis of blood serum samples.