Brain Tumor Segmentation In Mri Images Using Fully Convolved Neural Network With U Net Model

Document Type : Primary Research paper

Authors

1 Assistant Professor, Department Of Computer Science And Engineering, Bannari Amman Institute Of Technology, Sathyamangalam, Tamilnadu, India.

2 Assistant Professor, Department Of Computer Science And Engineering, Faculty Of Engineering & Technology, SRM Institute Of Science & Technology, Kattankulathur, Tamilnadu, India.

Abstract

Extracting The Tumor Affected Area Is A Crucial Task Faced While
Diagnosing. As The Mortality Rate Is Higher In Glioma, It Is Necessary To Segment The
Region Accurately. Magnetic Resonance Imaging (MRI) Is Opted For Examining The
Tumors. Data Produced By MRI Imaging Will Be In High Volume Which Consumes A
Lot Of Time For Segmenting Manually. Automatic Segmentation Helps In Earlier
Identification Of Affected Areas. Structural Variations In Brain Tumors Bring Difficulties
In Automatic Segmentation. In Recent Years, Deep Learning Based Algorithms Are Mostly
Preferred For Segmentation. Deep Learning Is Mostly Preferred As It Is Easy To Extract
The Features. U Net Based Convolution Neural Network Architecture Is Used Here For
The Segmentation Of Brain Tumors. U Net Architecture Has Up Sampling And Down
Sampling Which Are Symmetric In Structure. The Number Of Convolution Layers,
Pooling Layers And Relu Layers Are Used Based On Our Need. Here 5 Layers Are Used
In Which Each Layer Has Two Convolution Layers And A Pooling Layer. The Algorithm
Is Tested With The Brats Data Set (Brain Tumor Segmentation Challenge). The Proposed
Architecture Achieves The Accuracy Of 96.78%, Sensitivity Of 89.93% And Specificity Of
98.37%. With The Results, We Can Find That The U Net Architecture Helps In Exact
Segmentation In MRI Images Compared To All DCNN Algorithms.

Keywords