Abstract: The project is intended to classify the Brain MRI using the computer vision technique of Deep Learning. Brain MRI can be of two major types depending on the way they were extracted from the scanner as well as on the type of scanner and method used for taking the MRI of a subject. These two categories are T1weighted anatomical as well as T2weighted resting BOLD MRI images. These are categorized with the help of a Deep Neural Networks with 4 hidden layers. The dataset is taken from the openfmri. There are 120 MRI data and are released to the general public on as a part of the materials for “Temporal interpolation alters motion in fMRI scans: magnitude and consequence for artifacts detection” by Power et al. in PLOS ONE. Encased for each subject could be a T1-weighted anatomical picture (MP-RAGE) and one or extra T2*-weighted scans(resting bold outputs).The dataset we have is a 3D cuboid of the subject’s MRI image for the T1 weighted scans and 4D for the T2 weighted scans where the 1st, 2nd and 3rd being the x, y and z axis for the 3D image and the 4th dimension being the time. Every subject’s MRI is then split into 2D slices from all the axis to increase the data volume, then these images are pre-processed and fed into a 2D-CNN network. This is then trained for 3 epoch cycle in the cloud for a better processing speed and the resulted output of the weighted and biases are stored for the model to predict future inputs.
Keywords: MRI, BOLD, T1-weighted, T2-weighted, 2D-CNN.
|
DOI:
10.17148/IJARCCE.2020.9702
