Validation of a White Blood Cell Counting and Classifying Program of Demographic Factors Contributing to White Blood Cell Composition in Free Ranging Rhesus Macaques

Aging presents a complex array of challenges, including increased susceptibility to various diseases due to decrease in the effective function of the immune system. White blood cells, or WBCs, play a crucial role in providing insight into the state of the body and it’s immune system, and is thus, a vital biomarker. Traditionally, obtaining WBC counts involves many man hours and involves labor intensive hand counting of WBCs seen in a blood smear. To streamline this process, machine learning and artificial intelligence may be used. Using a cell counting program, or CCP, this thesis aims to validate the accuracy of the CCP’s capabilities in the cell counting process. We compared CCP generated WBC proportional counts with a ground truth data set, called Zooniverse. From this, a minimal to moderate correlation was found between the CCP generated data and the Zooniverse data. In conjunction with this, significant discrepancies were observed between certain WBC subtypes, suggesting limitations in the CCP performance. Further analysis of the CCP outputted data revealed an uneven distribution of age in the samples considered, which could have produced a biasing result. Linear model regression analyses using CCP data indicated few significant associations between age, sex and the resulting WBC proportions, casting further doubt on the program validity. Our findings highlight both the promise and limitations of automated WBC counting programs. While the CCP model in question depicted that it does indeed offer time saving benefits, the current model’s accuracy in capturing the subtle age related changes in WBC composition are not entirely confirmed. Future improvements in algorithm design and validation methods are necessary to enhance the use of this particular CCP.