A Comparative Study on Machine Learning Models for Automatic Classification of Cell Types from Digitally Reconstructed Neurons
Abstract: For the last decade, the use of machine learning in neuroscientific research has become a popular topic. For instance, image recognition has been used together with machine learning to detect and also help improve the diagnostics of diseases. This study compares the accuracy of a Convolutional Neural Network (CNN), a support vector classifier and a random forest classifier to investigate which are better suited for classification of cell types based on digitally reconstructed images from mice. All models were trained on both a larger unbalanced dataset containing 49 different cell types and a smaller balanced dataset containing only 3 types. Each model was evaluated on how accurate they could classify all cell types but also their accuracy on individual cell types. The results showed that the convolutional neural network had the best mean accuracy, with 51 and 83 percent on respective datasets. When looking at classification of individual cell types, all the models had good accuracy on at least a few cell types, but still, the CNN had the best individual accuracy and also consistency. In conclusion, the results showcase that a convolutional neural network is probably better suited when classifying cell types from digitally reconstructed images, but the other methods could also perform well on some of the cell types. However, further research is needed to reach a higher accuracy and reliability of the results.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)