Brain Tumor Detection Using Convolutional Neural Network with Dense Connections

Abstract

Radiologists examine various Magnetic Resonant Image (MRI) sequences produced using a multimodal imaging approach when treating a brain tumor. Research in neuro-oncology has recently concentrated on cutting-edge MRI methods that attempt to link histological characteristics to radiological phenotypes like cellularity or vascularity. The fundamental imaging procedure used in clinical practice today consists of T1-weighted sequences, which show anatomy, and T2-weighted sequences, which show oedema and can evaluate cellularity. Radiologists continue to attempt to diagnose a suspected glioma and a brain lesion on an MRI using the histological subtype. However, a significant problem with this diagnostic is that it is quite subjective. They only base the determination on qualitative, subjective characteristics such "low-to-moderate oedema could be reflecting the features of tumour," as accurate, quantitative thresholds have not yet been established. The diagnostic accuracy for finding the tumor is still a challenge; it is now moderate to low. A few years ago, AI made its way into neuro-oncology. AI algorithms can be utilized to segment and identify the subtypes of tumors in MR images in addition to detecting their presence. The goal of this study is to identify brain tumors using the Convolutional Neural Network (CNN) variation DenseNet. Under various architectural patterns, the performance of the pre-trained DenseNet and ResNet models were compared.