A comparative analysis of lightweight convolutional neural network models for brain tumor detection in MRI images with an explainable AI approach

Abstract

Accurate and timely detection of brain tumors from Magnetic Resonance Imaging (MRI) images plays a critical role in improving treatment outcomes and increasing patient survival rates. In recent years, models based on Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in medical image analysis; however, many of these models are computationally complex and thus unsuitable for practical applications, particularly in resource-constrained environments. In this study, a comparative analysis of several lightweight CNN models, including MobileNet, ShuffleNet, and EfficientNet, is presented for brain tumor detection from MRI images. A publicly available brain MRI dataset was utilized, and a preprocessing pipeline comprising normalization, data augmentation, and image quality enhancement was applied. The performance of the models was evaluated using metrics such as accuracy, sensitivity, specificity, and F1-score. In addition, to enhance model reliability, explainability techniques based on Grad-CAM were employed to analyze the regions influencing the decision-making process. The results indicate that lightweight CNN models, while significantly reducing computational complexity, are capable of achieving competitive accuracy in brain tumor detection. Furthermore, explainability analysis reveals that the models predominantly focus on tumor-relevant regions in most cases. The findings of this study can contribute to the development of fast, reliable, and clinically applicable diagnostic systems, particularly in resource-limited settings.