An intelligent brain tumor detection model using lightweight hybrid twin attentive pyramid convolutional network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early and accurate brain tumor (BT) detection is crucial for effective treatment and improved patient outcomes.
• Medical imaging, such as MRIs, provides valuable data but detecting brain cancers remains challenging.
• Deep learning approaches offer enhanced computational efficiency and precision for tumor classification.

Purpose of the Study:

• To propose a hybrid lightweight deep learning framework (HybLwDL) for advanced brain tumor diagnosis using MRI images.
• To enhance the accuracy and reliability of brain tumor detection through an innovative AI model.
• To improve the interpretability of AI-driven diagnostic tools by visualizing key image regions.

Main Methods:

• The HybLwDL framework utilizes a Gaussian Bilateral Network Filter (GANF) for MRI image pre-processing.
• A lightweight hybrid twin-attentive pyramid convolutional network (LHTA-PCNet) with ResNet backbone performs feature extraction and classification.
• Hyper-parameters are optimized using the Stellar Oscillation Optimizer (SOO), and Grad-CAM visualizes significant regions.

Main Results:

• The proposed HybLwDL model achieved a high classification accuracy of 99.5% on the BT Detection 2020 dataset.
• The LHTA-PCNet model effectively extracts local and global features and acquires multi-scale information.
• The integration of GANF and SOO contributed to improved image quality and classification performance.

Conclusions:

• The HybLwDL framework demonstrates high reliability and accuracy for brain tumor classification from MRI scans.
• This deep learning approach offers a promising tool for early and precise brain tumor detection.
• The visualization capabilities enhance the understanding of AI model decisions in medical diagnostics.