NestedVGG: A Novel Deep Learning-based Architecture for Brain Tumour Diagnosis

Brain tumors are sudden outbreak of cells in the brain or on its surface and are categorized into gliomas, meningiomas, and pituitary adenomas. Traditionally, tumors of the brain are categorized using a manual approach that includes visual examination by radiologists on Magnetic Resonance Imaging (MRI). This is arduous and subject to inter-expert variability. In view of the subtle and fine-grained compact variations that exist in MRI images of these three types of brain tumors, this paper proposes an automated brain tumor detection system using deep learning (DL) with MRI scans. This paper presents an automated system for brain tumor detection using DL on MRI scans. The "NestedVGG"model implements transfer learning(TL) with the outer architecture of "Outer VGG"for feature extraction and the inner architecture for fine-grained identification. This "Outer VGG"utilizes a pre-trained VGG16 model to acquire information about features from MRI images. While the inner "Inner VGG"can capture minute variations between tumor types to classify glioma, meningioma, no-tumour, and pituitary cases effectively. Publicly available MRI dataset consists of three classes of tumours were used for testing and this framework has achieves accuracy of 97.71% on the test set, thus outperforming the prior methods in this task. Moreover, it has maintained its high accuracy across various types of tumors, thus making the model flexible. This paper furthers brain tumor diagnosis through an effective architecture for brain tumor identification, which can lead to advanced diagnostic tools and better patient care.