Support vector machine? – not only for MRI-images

Introduction Nowadays, the plethora of scientific data produced way exceeds the humanʼs brain capacity, so automatic recognition tools may be used to avoid information loss and speed up data processing. Techniques like machine learning or deep learning are common in MRI-diagnostics, but their algorithms are also able to simplify the analysis of microscopic pictures.

Methods Using mentioned principle, Support Vector Machine (SVM) was developed as a part of the project “Occurrence and Spreading of TMEM-119-expriming Microglia in Schizophrenia and Affective Disorders”. The tool is able to differentiate between microglia (TMEM-119-positive) and capillaries (Factor-VIII-positive), which can be used to study interactions between microglia and blood-brain barrier and measure distances between microglia and other microscopic objects.

Results The SVM algorithm uses four parameters, which accurately describe the morphology of microglia cells, including itʼs ramified and amoeboid form, for differing objects. These parameters can be calculated manually or received by image recognition and analysis tools like MatLab (3DMorph (York et al., 2014)) or Python.

Conclusion The performance of the tool was trained on 5000 samples of human microglial cells and 2000 samples of human brain capillaries. The sensitivity of recognition in test data set was 85% and 82% for microglia and capillary respectively, the specificity was 78% for both objects. The SVM operation was improved by increasing the variety of different microglia staining, inclusive HLA-DR-staining and TMEM-119-staining and by including data from rat brain tissue.