BrainMRNet: Brain tumor detection using magnetic resonance images with a novel convolutional neural network model

Area of Science:

• Medical Imaging and Artificial Intelligence
• Biomedical Engineering
• Neurology

Background:

• Brain tumors are abnormal growths affecting human life, often treated surgically.
• Deep learning models are increasingly utilized for disease diagnosis and treatment in biomedicine.

Purpose of the Study:

• To introduce BrainMRNet, a novel convolutional neural network for brain tumor detection.
• To evaluate BrainMRNet's performance against established deep learning models.

Main Methods:

• BrainMRNet incorporates attention modules and hypercolumn techniques within a residual network architecture.
• Image preprocessing and augmentation are applied before feature extraction via attention modules.
• The hypercolumn technique retains features from all layers in the final layer for optimal selection.

Main Results:

• BrainMRNet demonstrated superior performance compared to pre-trained models like AlexNet, GoogleNet, and VGG-16.
• The proposed model achieved a high classification accuracy of 96.05% for brain tumor detection.
• Magnetic resonance images were utilized for model training and validation.

Conclusions:

• BrainMRNet presents a highly successful and efficient deep learning approach for brain tumor detection.
• The integration of attention modules and hypercolumn techniques enhances feature selection and diagnostic accuracy.
• This model offers a promising advancement in the application of artificial intelligence in neuro-oncology.