Integrating MobileNetV3 and SqueezeNet for Multi-class Brain Tumor Classification

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Neuroscience

Background:

• Accurate brain tumor classification is crucial for effective treatment.
• Traditional MRI analysis is time-consuming and subjective.
• Automated classification methods are needed to improve efficiency and consistency.

Purpose of the Study:

• To evaluate lightweight deep learning models for multi-class brain tumor classification.
• To compare the performance of MobileNetV3, SqueezeNet, and a hybrid model.
• To assess the trade-off between accuracy and computational efficiency.

Main Methods:

• Utilized a dataset of 7023 MRI images for four categories: glioma, meningioma, pituitary tumors, and no tumor.
• Investigated individual and feature-fused MobileNetV3 and SqueezeNet models.
• Employed Grad-CAM for model interpretability and visualization.

Main Results:

• MobileNetV3 achieved the highest test accuracy of 99.31% with only 3.47M parameters.
• The proposed lightweight models outperformed baseline architectures like VGG16 and InceptionV3.
• Grad-CAM visualizations effectively highlighted tumor-specific regions.

Conclusions:

• Lightweight deep learning models, particularly MobileNetV3, offer a highly accurate and computationally efficient solution for brain tumor classification.
• These models demonstrate potential for real-world clinical deployment.
• Optimized lightweight networks provide interpretable and accurate diagnostic tools.