Comparison of automated and manual approaches for microglial quantification and classification: A focus on the HALO...

Phenotypic changes in microglia have been linked to multiple neurological conditions, such as dementia, Parkinson's disease, stroke and traumatic brain injury. Consistent identification and classification of microglia is essential in understanding potential links with neurological diseases. Currently, there are several ways by which the microglial population and morphology are assessed, including manually or using open-source image analysis platforms, such as ImageJ. A microglial classification module for the HALO digital pathology platform has been developed for this purpose but has not yet been validated within the literature. The current study therefore conducted a comparison of the performance of this HALO module to manual microglial analysis and to automated analysis via ImageJ using both human and rat brain tissue. In 5 μm thick human tissue, total and activated microglia/mm² counted by HALO showed strong positive correlations with both manual and ImageJ counts. HALO did not differ from the other methods for total microglia counts; however, Halo did differ from both manual and ImageJ methods in the number of activated microglia detected within the substantia nigra. In 20 μm rat tissue, total counts derived from HALO showed moderate positive correlations with both manual and ImageJ counting; however, activated counts on Halo were not positively correlated with any method. To our knowledge, this is the first study to systematically compare the Halo module to other common methods of microglia analysis. When applied to 5 μm tissue, the Halo module is comparable to manual counting and to automated analysis on ImageJ. However, when analyzing thicker tissue, Halo struggles to perform in line with these other methods, particularly for counts of activated microglia, likely due to increased cell density and the morphological complexity of microglia. These results highlight the importance of carefully tailoring image analysis parameters on automated counting methods to suit the needs of the tissue.