Brain Tumor Classification in MRI Scans Using Edge Computing and a Shallow Attention-Guided CNN.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Brain tumors necessitate accurate Magnetic Resonance Imaging (MRI) for diagnosis and treatment.
• Deep learning (AI) shows promise in MRI analysis but often results in large, slow models unsuitable for edge computing.
• Efficient AI models are crucial for practical clinical deployment in brain tumor classification.

Purpose of the Study:

• To develop a novel, lightweight attention-guided classification model for brain tumors.
• To investigate methods for reducing model parameters without compromising diagnostic accuracy.
• To enhance the practicality of AI-driven brain tumor analysis for medical edge computing.

Main Methods:

• A shallow attention-guided convolutional neural network (ANSA_Ensemble) was developed.
• Model compression techniques and Monte Carlo simulations were employed for evaluation.
• The model was validated on three diverse, open-source brain tumor datasets.

Main Results:

• The ANSA_Ensemble model achieved high accuracies: 98.04% (best) and 96.69% (average) on the Cheng dataset.
• Comparable results were obtained on the Bhuvaji (95.16%) and Sherif (95.20%) datasets.
• Depthwise separable convolutions offered the optimal balance between accuracy and speed.

Conclusions:

• The proposed attention-guided model demonstrates performance on par with state-of-the-art methods.
• Increasing attention blocks consistently improved model accuracy.
• The study provides an efficient AI solution for brain tumor classification, with publicly available code.