DCNNBT: A NOVEL DEEP CONVOLUTION NEURAL NETWORK-BASED BRAIN TUMOR CLASSIFICATION MODEL

Abstract

An early brain tumor diagnosis is crucial for effective and proactive treatment, which improves the patient's survival rate. In this paper, we propose a novel Deep Convolutional Neural Network for Brain Tumor (DCNNBT), which detects and classifies brain tumors. The key differentiators of this paper are dimension scaling for image resolution, depth of layers, and width of channels with rigorous optimization of the hyperparameters. DCNNBT classifies and detects four types of brain tumors: benign, pituitary, glioma, and meningioma based on axial, coronal, and sagittal-coronal views. The DCNNBT was developed and tested on two public MRI datasets with more than 403,064 images containing four modalities for 872 patients. The performance of DCNNBT was evaluated against six well-established pre-trained deep learning (DL) models, including SE-ResNet-101, SE-ResNet-152, SENet-154, ResNet152V2, EfficientNetB0, and EfficientNetB5, through transfer learning. In the comparison, DCNNBT showed high accuracy of 99.18% for brain tumor classification, significantly higher than the other studies based on the same database.