BRAIN TUMOR DIAGNOSIS USING CNN WITH HYBRID OPTIMIZER

Abstract

Brain tumors are recognized as severe illnesses wherein the precision of images plays a crucial role. Accurately identifying tumors is essential for precisely pinpointing the affected area and thereby reducing the mortality rate. Consequently, understanding the hidden patterns becomes imperative for an improved diagnosis and image quality. However, achieving accurate diagnoses across various lesion cases poses a significant challenge. To address the limitations of existing methods, proposes a comprehensive approach, initially employing the k-nearest Neighbors (KNN) classifier and subsequently transitioning to a Convolutional Neural Network (CNN) for enhanced performance. For the KNN classifier, data normalization and reshaping ensure compatibility, followed by training and prediction. However, due to the limitations of the KNN classifier such as sensitivity to noise and feature complexity, KNN classifier performance is low. So CNN Classifier is used. The CNN architecture integrates convolutional layers for feature extraction, ReLU activation for non-linearity, max-pooling for spatial reduction, flattening for dense layer input, and dense layers for classification. The hybrid (Adam and SGD) optimizer is employed to refine the performance of the Convolutional Neural Network (CNN), enhancing its ability to accurately extract features and classify brain tumor images from MRI scans. The results are expected to show that CNN performs better than kNN indetecting brain tumors from MRI images, highlighting the effectiveness of advanced deep-learning methods for more accurate medical image analysis.