Tumor Identification and Classification of MRI Brain Images Using Deep Learning Optimizers (Adam vs Sgdm)

Diagnosing brain tumors presents a complex and challenging task for medical professionals. It requires accurate detection and classification of tumor volume, which is essential for determining the appropriate treatment. In recent times, deep learning algorithms, such as Convolutional Neural Networks (CNNs), have shown the best results in medical image analysis. For our study, we utilized a CNN-based method to detect brain tumors by analyzing MRI scans. The use of Deep Learning for segmenting and classifying tumors presents an excellent improvement nowadays both in identification and classification. In this project, we studied the application of two popular deep learning optimizers namely Adam (Adaptive Moment Estimation) and SGD with Momentum (Sgdm), to improve the accuracy and efficiency of tumor identification and classification from MRI brain images By comparing the performance of these optimizers, we aim to elucidate their effectiveness in the context of medical image analysis. Here the classification of BTs into 4 classes (Normal, Pituitary Tumor, Meningioma Tumor, Glioma Tumor). For training purposes, T1-weighted images are used. The testing accuracies for both Deep Learning optimizers achieve 98-99 % with variation in time and learning parameters. The proposed algorithm signifies the classification outcomes of many state-of-art-methods.

I. Introduction

Imagine a bustling city, which represents the brain, with its complex network of streets and alleys, where every thought and movement is orchestrated seamlessly. In this bustling city, imagine an unexpected visitor - unseen, silent, and potentially disruptive. This visitor is a brain tumor, which is a group of cells growing uncontrollably and disrupting the harmony of the city. Fig 1.

Brain analysis