Clustering functional mri data using a robust unsupervised learning algorithm

Functional Magnetic Resonance Imaging (fMRI) has provided neuroscientists with a powerful tool to examine brain activity by calculating the levels of oxygen in the blood and generates a sequence of 3-D images. Clustering approach is a model-free analysis; it has the ability of defining the active zones in the brain without the need of prior knowledge about activation patterns or experiment as the classical and statistical General Linear Model (GLM) method. The goal of this proposal is to find a solution for choosing an appropriate clustering approach to obtain the best performance for whole brain functional connectivity by means of data analysis. In this work, a novel and robust unsupervised learning approach is proposed; it relies on using a Robust Growing Neural Gas (RGNG) algorithm into a real auditory fMRI dataset. The main contribution of this work is running the RGNG algorithm into fMRI dataset with a comparison to NG and GNG algorithms, which is not used for the purpose yet or any other applications also. Another comparison has been done with the model-based data analysis approach using a Statistical Parametric Mapping (SPM) package which is based on GLM. The output results demonstrate that the presented RGNG approach is clearly superior to other approaches as revealed by their performance measured by Minimum Description Length (MDL) and Receiver Operating Characteristic (ROC) analysis. A MATLAB-based graphical user interface (GUI) tool is designed and implemented as a software package for fMRI data analysis. A new model for neuroscience data analysis is developed in this work which is easily accessible by researchers. The proposed work could help the neurologist and psychologist for better interpretation of fMRI dataset.