Brain Tumor Detection Using Deep Learning

Abstract

Recent studies on utilizing deep learning (DL) algorithms for the identification of brain tumors have focused primarily on gliomas (tumors arising from glial cells), meningiomas (tumors of the meninges), and pituitary tumor types (tumors of the pituitary gland) with a variety of different approaches. Convolutional neural networks (CNNs) and hybrid architectures will be investigated in the detection and classification of tumors within magnetic resonance imaging (MRI) scans at an early stage of disease progression. The proposed work will utilize the concepts of transfer learning by employing pretrained CNN architectures, such as YOLOv7, VGG16, and ResNet as well as methods for enhancing DL accuracy, such as convolutional block attention modules (CBAM), and explainable artificial intelligence (EAI) techniques (such as Grad- CAM) as tools to achieve high degrees of tumor detection accuracy (>95- 99%) and generate clinically applicable heat maps that can assist clinicians in performing diagnostic tasks. Following the detection of a tumor, each individual will receive a tailored evidence-based precaution model to assist them in monitoring their health by developing personalized recommendations that will outline how to track their symptoms based on their specific type of tumor (e.g., symptom tracking for gliomas, imaging frequency for meningiomas, hormonal testing for pituitary tumors).

Key methodologies utilized in the study include the following:

• DL automates the extraction of features from MRIs while providing a diagnostic support (DS) solution to facilitate treatment planning following tumor


• The researchers developed multiple different types of CNNs combined with the YOLO model for the purpose of developing a model for tumor skin identification while also developing an ensemble model using machine learning or rule-based approaches based on the specific tumor classification to generate precautionary recommendations based on the characteristics of the tumor.