NeuroNet: An Attention-Driven Lightweight Deep Learning Model for Improved Brain Cancer Diagnosis

This paper introduces NeuroNet architecture, a lightweight deep-learning framework designed for brain tumor identification, integrating a spatial attention-driven convolutional neural network (CNN) architecture. NeuroNet aims to enhance the classification accuracy of brain tumors, specifically targeting three types: gliomas, meningiomas, and general brain tumors, using Magnetic resonance imaging (MRI) data. The framework is estimated on the Bangladesh Brain Cancer MRI Dataset, which consists of 6,056 MRI images divided into three balanced classes. Unlike traditional machine learning techniques that rely on handcrafted features, NeuroNet's architecture leverages CNN layers with a spatial attention mechanism, enabling the model to focus on critical tumor regions. This approach reduces manual intervention and enhances feature representation. The proposed model performed a high classification accuracy of 99.04%, with a weighted F1-score of 99.02%, outperforming conventional models. These results indicate NeuroNet's potential to assist medical professionals by providing an accurate and reliable diagnostic tool, thus improving brain tumor treatment outcomes. This research underscores the promise of attention mechanisms within deep learning frameworks for advancing cancer diagnosis and medical imaging.