A Study on the Classification of Brain MRI Images for Brain Tumor Detection

A Comparative Analysis

Abstract

Accurate classification of brain tumors is essential for effective diagnosis, as it directly affects treatment choices, patient outcomes, and survival rates. Early and precise detection enables timely interventions, reducing the risk of tumor progression. Without reliable classifiers, traditional feature extraction techniques have drawbacks. Our study suggests a hybrid model that incorporates the advantages of Random Forest Classifier (RFC), Support Vector Machine (SVM), and Visual Geometry Group (VGG16) to improve classification performance. To improve feature extraction, the Magnetic resonance imaging (MRI) images are pre-processed. This includes pixel intensity. To improve data diversity, augmentation techniques such as random flip, random height, random width, horizontal flip, and vertical flip are used. Next, a unique Deep Convolutional Neural Networks (DCNN) that is VGG16 is created to extract significant deep features. Evaluation of the model's performance using various optimizers revealed that the RMSprop optimizer outperformed models employing Adam (80.39%) and SGD (64.71%), achieving the highest validation accuracy (82.35%). SVM obtained a validation accuracy of 47.06%, while RFC obtained 64.71%. These results show the importance of classifier and optimizer selection. This study highlights the efficacy of the VGG16 model with the RMSprop optimizer and shows the potential of integrating deep learning and conventional machine learning approaches for brain tumor classification. It demonstrates the potential of combining deep learning and traditional machine learning techniques for brain tumor classification, highlighting effectiveness of the VGG16 model with the RMSprop optimizer while emphasizing the need for further exploration of optimizers and classifiers to enhance overall model performance and robustness.