Noise Reduction in Brain MRI Images: A Gaussian Filtering Approach to Enhance Deep Learning Models for Classification

Brain tumors pose a significant health risk and are a leading cause of mortality globally, with the World Health Organization (WHO) projecting an annual increase of 5% in their incidence. This study aims to analyze the performance of deep learning methods in identifying different types of brain tumors, emphasizing the importance of timely diagnosis. The comprehensive methodology presented in this study serves as a systematic guide for implementing and evaluating brain tumor classification models using deep learning and machine learning techniques. The research demonstrates the effectiveness of using pre-trained models on MRI datasets, including VGG-16, ResNet-50, and AlexNet, to achieve accurate classifications. The impact of noise during image acquisition via MRI has also been considered by applying noise removal techniques. Implementing these methods has resulted in improved model accuracy and efficiency. The evaluation framework incorporates essential metrics such as confusion matrices, accuracy, F1-score, and MCC coefficient for each approach. Among the proposed algorithms, ResNet-50 achieved the highest accuracy of 89.683% with the LDA classifier when no image enhancement was applied. When image enhancement was employed, ResNet-50's accuracy increased to 90.296% with the LDA classifier. Conversely, in the case of AlexNet, accuracy decreased when noise removal was applied. Also, VGG-16 showed an increase in accuracy with the SVM classifier when image enhancement was applied.