Brain tumor segmentation by dilated convolution neural network in MRI images

homayoon yektaei¹ mehrnoosh bahmani² hanieh yektaei³

1) master of biomedical engineering, tehran shomal
2) master of biomedical engineering, tehran shomal
3) bachlor of biomedical engineerin, ahrar

IOI Code : XBGE-FCEEG
Publication : 11th International Conference on New Solutions in Engineering, Information Science and Technology of the Century ahead(icietconf.com)
Abstract :
    Nowadays, due to the increase in the number of patients with brain tumors, diagnosis and isolation of the tumor plays an important role in the process of treatment and surgery. Due to the high error of manual segmentation of the tumor, algorithms that perform this operation with less error are of great importance. Convolutional neural networks have made great progress in the field of medical imaging. In this study, different structures of convolution networks using the capability of dilated Convolution layers for glioma tumor segmentation in brain tumor segmentation challenge MRI images are presented. In the first structure, the layers of the Convolution are aligned in parallel and then interconnected. In the other two structures, the dilated convolutional layer with different number of layers is used. This layer is capable of preserving the size and resolution of the output image. It also does not require additional effort to extract smaller images from the dataset. Dilated convoluted capability has been used to maintain output image resolution, lower memory consumption and shorter runtime. Due to the capability of dilated cannulation layers and fully cannulated structure, the network has been able to maintain the output size and prevent the training of additional parameters, which has resulted in higher precision than other structures, and the duration Program execution and memory usage are reduced compared to other structures. This method also increases the accuracy of diagnosis.

Keywords : segmentation brain tumor neural network dilated convolution mri