Brain Tumor Classification in MRI Images Using Convolutional Neural Networks with Explainable Artificial Intelligence
TL;DR
This study aims to enhance the transparency of Convolutional Neural Network (CNN)-based brain tumor classification models by implementing Explainable Artificial Intelligence (XAI) techniques, specifically Eigen-CAM and LIME, utilizing a dataset of 3,000 MRI images.
Abstract
Brain tumors are a condition requiring rapid and accurate diagnosis. This study aims to enhance the transparency of Convolutional Neural Network (CNN)-based brain tumor classification models by implementing Explainable Artificial Intelligence (XAI) techniques, specifically Eigen-CAM and LIME. The models were developed using a transfer learning approach on ResNet50 and EfficientNetB0 architectures, utilizing a dataset of 3,000 MRI images categorized into glioma, meningioma, and pituitary tumor classes. Test results indicate that ResNet50 achieved the best performance, with accuracy, precision, recall, and F1-score values of 94%, while EfficientNetB0 achieved 93%. The application of 5-fold cross-validation improved the models' generalization capabilities and reduced the risk of overfitting. Visualizations using Eigen-CAM and LIME demonstrate that the models focus on relevant tumor regions, thereby increasing the transparency and reliability of MRI-based classification.