Brain tumor classification with convolutional neural network

EasyChair Preprint 5988, version 1

Abstract

Brain tumor is an acute cancerous disease that results from abnormal and uncontrollable cell division. Brain tumors are classified by biopsy, which is not usually performed before final surgery of the brain. Recent advances and improvements in technology in deep learning have helped the health industry in medical imaging for the medical diagnosis of many diseases. A deep learning algorithm that has yielded substantial results in image segmentation and classification is the Convolutional Neural Network (CNN). Similarly, in our paper, we introduce the convolutional neural network (CNN) approach along with Image Processing to categorize brain MRI scan images into four classes which are glioma tumor, meningioma tumor, pituitary tumor and normal patients. Using the transfer learning approach and a CNN based model from scratch we compared the performance of our scratched CNN model with pre-trained inceptionresnetv2, inceptionv3 models. The experiment is tested on a brain tumor MRI dataset which contains (826) MRI images for glioma tumor patients, (822) MRI images for meningioma tumor patients, (827) MRI images for pituitary tumor patients and (835) MRI images for normal persons. but the experimental result shows that our model accuracy result is very effective and have very low complexity rate by achieving 93.15% accuracy for the transfer learning model, while inceptionresnetv2 achieved 86.80%, inceptionv3 achieved 85.34% and BRAIN-TUMOR-net based CNN achieved 91.24% accuracy. Our model requires very less computational power and has much better accuracy results as compared to other pre-trained models.

Keyphrases: Brain Tumor, Medical Imaging, deep learning

@booklet{EasyChair:5988,
  author    = {Mohamed Shoaib and Mohamed Elshamy and Taha Taha and Adel El-Fishawy and Fathi Abd El-Samie},
  title     = {Brain tumor classification with convolutional neural network},
  howpublished = {EasyChair Preprint 5988},
  year      = {EasyChair, 2021}}