Indian Journal of Public Health Research & Development

Despite advantages in computer vision and machine learning its application in classification of Brain tumors and achieving optimal results is yet remains a challenge. The tumors appear at any location in the brain by nature and the tumors have any kind of dimensions, contour and contrast. This sources the motivation of our investigation of machine learning that adventures a high capacity deep learning algorithm being enormously effectual.

Deep learning has been used successfully in supervised classification tasks in order to learn complex patterns. The purpose of this research is to implement the machine learning technique to classify the tumors in the brain, with different classes of tumors such as benign and malignant. Training neural networks over the dataset is taken from the open f-MRI. There are 120 MRI datasets that are released to the public along as part of the materials for “Temporal interpolation alters motion in fMRI scans: magnitude and consequence for artifacts detection” Included for each subject is a T1-weighted anatomical image (MP-RAGE) and one or more T2weighted scans (resting BOLD scans), legacy.openfmri.org. Every subject's MRI is then split into 2D slices from the entire axis to increase the data volume, and then these images are preprocessed and fed into a 2D-CNN network. It is then trained for number of epoch cycle for a better processing speed and the resulted output of the weighted and biases are stored for the model to predict future inputs this has verified to be precise in its classifications with an average five-fold cross validation of 91.43%.