Abstract:
Segmentation of brain tumor images is one of the most essential and challenging
tasks in the field of medical image processing, as human-assisted
manual categorization can lead to erroneous prognosis and prediction. In
addition, it is a difficult process when there is an abundance of data available
to assist. The difficulty of extracting brain tumor regions from MRI images
due to the wide diversity of brain tumor appearances and their similarity
to normal tissues. In this paper, we propose a modified U-Net architecture
within a deep learning framework for the detection and segmentation of
brain lesions in MRI images. On authentic images from the Medical Image
Computing and Computer-Assisted Interventions UW-Madison GI Tract
Image Segmentation datasets, the applied model has been evaluated. Using
the above-mentioned dataset, a test accuracy of 99.4% was attained. A
comparison with other publications demonstrates that our U-Net-based
model outperforms other deep learning-based models.
