Improving MRI Classification through Layered Convolutional Neural Networks Configuration

Timely and accurate classification of brain tumors using Magnetic Resonance Imaging (MRI) is critical for effective treatment planning. This study proposes a layered Convolutional Neural Network (CNN) configuration to enhance the classification of brain tumors, addressing the limitations of traditional machine learning approaches that rely heavily on manual feature extraction. Utilizing a dataset sourced from Kaggle, comprising 7023 MRI images categorized into glioma, meningioma, no tumor, and pituitary tumor classes, the research implements data augmentation techniques such as rotation and flipping to increase the dataset size by 20%. Images were standardized to 128x128 pixels and normalized for model compatibility. The core model architecture was built using 2D CNNs with configurations ranging from one to three layers. The models were trained and tested using TensorFlow and Keras on Google Collaboratory, and evaluated based on accuracy, loss, and computational efficiency. The findings revealed that among all the configurations tested, the three-layered CNN model delivered the best performance. It achieved an accuracy value of 89.79% with a corresponding loss of 0.469. In terms of processing time, the model completed training in 59.8894 seconds and performed inference in 5.1099 seconds, highlighting its suitability for real-time diagnostic applications despite the longer training duration.