BrainNet: Optimal Deep Learning Feature Fusion for Brain Tumor Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Early detection of brain tumors is critical for patient survival.
• Automated classification of brain tumors using medical imaging (FLAIR, T1, T2, T1CE) aids diagnosis.
• Deep learning models like ResNet101 show promise but can generate redundant features, increasing computational load.

Purpose of the Study:

• To develop a fully automated design for brain tumor classification.
• To optimize deep learning features for improved accuracy and reduced computational overhead.
• To enhance the efficiency of brain tumor detection in medical imaging.

Main Methods:

• Utilized ResNet101 for transfer learning on brain tumor datasets (FLAIR, T1, T2, T1CE).
• Employed differential evolution and particle swarm optimization to identify optimal deep learning features.
• Applied serial fusion and Principal Component Analysis (PCA) for feature vector optimization.
• Integrated optimized features with various classifiers for tumor classification.

Main Results:

• Achieved a prediction time speedup of 25.5x on a medium neural network.
• Attained a classification accuracy of 94.4% for brain tumors.
• Demonstrated significant reduction in computational overhead compared to state-of-the-art methods while preserving accuracy.

Conclusions:

• The proposed automated method effectively classifies brain tumors using optimized deep learning features.
• The technique offers a substantial improvement in computational efficiency for brain tumor detection.
• This approach holds potential for faster and more efficient clinical diagnosis of brain tumors.