Robust Brain Tumor Classification of MRI Images Using Deep Learning via InceptionResNetV2

The increasing application of AI (Artificial Intelligence) and ML (Machine Learning) in various domains has shown great promise, particularly in addressing complex real-time challenges. One of the most impactful areas is DL (Deep Learning), which has proven essential in medical image analysis. This paper proposes a robust method for classifying brain MRI images using InceptionResNetV2 architecture. The dataset is preprocessed by converting grayscale images to RGB and applying an extensive data augmentation pipeline, including random flipping, rotation, contrast adjustment, zooming, and translation. The model is enhanced with additional classification layers and optimized using Adam optimizer. Training is improved with callbacks such as ReduceLROnPlateau and a custom ReduceLROnMultipleAccuracies for dynamic learning rate adjustment. The proposed model, trained for 30 epochs, achieved a validation accuracy of 99.45%. Class-wise metrics indicated high precision, recall, and F1-scores across all classes, indicating the model's robustness and effectiveness in classifying brain MRI images, making it a valuable tool for aiding in the diagnosis of neurological conditions.