Abstract:
The automation of brain tumor segmentation from MRI is an active topic in
the field of medical research. Different approaches and methods are being
proposed throughout the years to address this challenging task. The
application of convolutional neural network has caught the attention of many
researchers for the solution of this particular problem due to its extraordinary
performances in the field of computer vision . Many of the state-of-the-art
techniques use different approaches based on CNN. One of such approaches
is the multi-path CNN architecture.
In this paper, we propose a novel multi-path CNN architecture that allows
flow of information in two different pathways resulting in the exploitation of
both local and global features simultaneously, hence this architecture can
also be called two-pathway architecture. We use this architecture to train on
the dataset collected from BRATS 2013 challenge. Our model when tested on
BRATS 2013 training images, showed on an average 95 .654% accuracy.
