Brain fMRI Image Classification and Statistical Representation of Visual Objects

摘要

The application of the given paper work is to estimate what image a human brain is visually perceiving based on the neuroimaging information observed from the ventral temporal cortex (VT) portion. In the process, we used the nilearn library from python repository along with the haxby dataset which includes a set of functional MRI from 6 subjects viewing images that contains a grid of black and white pictures of some certain objects. Firstly, the haxby dataset was collected and few pre-processing steps such as masking, scaling and smoothing was done in order to reduce the complexity, noise and to standardize the data. Furthermore, the entire dataset was splitted into 80% of training example and 20% of test example. After that, the training examples were passed through a set of machine learning frameworks which consist of `Nearest Neighbors', `Linear SVM', `RBF SVM', `Gaussian Process', `Decision Tree', `Random Forest', `Neural Net', `Ada-Boost', `Naive Bayes' and `QDA' algorithms. Completing the training, the accuracy of the frameworks were tested and on an average the most accuracy of 95% was found with Neural Network and Support Vector Machine (SVM) across all the subjects.