Medical Image Analysis Using GAN-Based Augmentation and EfficientNet-B7 for Accurate Lesion Classification

Abstract

Brain tumors are caused by the uncontrolled and rapid growth of brain cells, posing a serious health threat if not detected early. Despite promising advancements, accurate identification and classification of brain tumors remain difficult. The number of cases is increasing globally, leading to thousands of deaths each year. Therefore, precise diagnosis is critical for effective treatment. Traditional machine learning (ML) techniques rely on manually designed features, which require significant effort. In contrast, deep learning (DL) has gained popularity for its ability to automatically extract features, making it highly effective for brain tumor detection and classification. This study proposes a novel framework that leverages a Generative Adversarial Network (GAN) to augment medical image datasets, thereby mitigating overfitting and enhancing model robustness. To further improve image quality, Contrast Limited Adaptive Histogram Equalization (CLAHE) is employed to enhance contrast and highlight important features. The enhanced images are then classified using the EfficientNet-B7 architecture, known for its high accuracy in medical imaging tasks.A total of 3,804 images were used in this study. The proposed model achieved an impressive classification accuracy of 98.27%. Compared to existing models, this approach demonstrated superior performance and reliability, making it a promising solution for the early detection and classification of various types of lesions.