A Machine Learning Approach for Brain Tissue Recognition of Human Brain Slice Images

This work presents a Machine Learning (ML) approach for classifying areas of brain tissue in a stack of high resolution human brain slice images. Compared with standard image segmentations algorithms, this ML approach provides more reliable results by concentrating on pixel classification. The presented ML approach is fourfold. First, four feature extraction methods were developed to extract features as a basis for the classification procedure. Second, two feature selection approaches were developed and implemented in order to construct feature vectors. Third, Random Forest (RF), Neuronal Networks (NN), and a novel ensemble Meta classifier constructed by different multilayer perceptron (MLP) were implemented in our classifier construction procedure. Finally, a post-processing method based on graph cut algorithm was used to enforce a smoother classification result into coherent regions. This paper details the feature extraction part and illustrates its application conceiving initial results of a small subset of brain slices.