Immunohistochemical Verification of Synapses and the Use of Machine Learning for the Analysis of the Synaptoarchitecture of the CA3 Hippocampus in White Rats During the Post-Ischemic Period

This study examines the methodological aspects of applying machine learning algorithms to explore new opportunities in interpreting inter-neuronal connections. The aim was to demonstrate that the combination of Ilastik and StarDist is effective for the morphometric characterization of giant synaptic terminals in the stratum lucidum of CA3 in the hippocampus of white rats under normal conditions and in the post-ischemic period. Material and methods . Cerebral ischemia in Wistar white rats was modeled by bilateral occlusion of the common carotid arteries (OCCA) for 20 minutes. Animals were studied without intervention (n=6, control) and at 6 hours, 1, 3, 7, 14, and 30 days after OCCA (n=36). Staining with hematoxylin and eosin, Nissl staining with thionine, as well as immunohistochemical reaction for synaptophysin, were used. Numerical density (NDT), sizes, staining intensity, and area of the terminals were determined, and the Ilastik and StarDist plugins were applied on the ImageJ/Fiji platform. Statistical analysis was performed using non-parametric methods in Statistica 8.0. Results . The relative area of the terminals did not differ between the manual method and machine learning. Machine learning provided additional information on numerical density, sizes, and average brightness of the terminals. At 6 hours after OCCA, NDT decreased by 44.3%, but then recovered over 7 days. The average area of the terminals was 16.7% larger at 6 hours and 1 day, but smaller than the control level at 14 days. The brightness of the terminal pixels was inversely proportional to the content of chromogen: it increased at 6 hours and 1 day after OCCA, then returned to control levels. Correlations were observed between the area and brightness of the terminals (R=0.78). Conclusion . The use of the combination of Ilastik and StarDist allowed for accurate assessment of numerical density, sizes, shape, relative area, and staining intensity of synaptic terminals in the hippocampus. Compared to the manual method, the application of machine learning provided significantly more information about the terminals in color immunohistochemical images.