Application of different classification techniques on brain morphological data

The increasing number of people affected by Neurodegenerative diseases and the improvement of brain imaging diagnostic techniques are bringing to a massive production of brain images that need demanding preprocessing and analysis algorithms. We analyzed volumetric measures of critical brain areas by using different Data Mining methods. Structural magnetic resonance images, generated in our university, were preprocessed using a fully automated segmentation method and the extracted volumetric information was then analyzed by using different binary classifiers. We performed three binary classification experiments considering different data mining algorithms and neurological diseases. Naïve Bayes outperformed all the others classifiers in two experiments, obtaining respectively 93.75% and 95.00% accuracy, while in the third experiment the best classifier was SVM but with a lower accuracy (58,56%). Afterwards, using the Stacking technique we combined the predictions from the best detected three models to build a meta-learner. Meta-learner classification results suggest that the application of the Stacking technique needs more experimentation and the test of additional stackers.