FusionNet: Dual input feature fusion network with ensemble based filter feature selection for enhanced brain tumor...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Brain tumors present a significant health challenge, necessitating accurate and rapid diagnosis for effective treatment.
• Traditional diagnostic methods face limitations in achieving high accuracy due to the complexity of brain tumors.
• Deep learning shows promise in automating brain tumor classification from MRI images, potentially enhancing diagnostic accuracy.

Purpose of the Study:

• To introduce FusionNet, a novel deep learning approach for enhanced brain tumor classification.
• To improve diagnostic accuracy by utilizing both normal and segmented MRI images.
• To advance the field of automated brain tumor detection and classification.

Main Methods:

• Generating segmented MRI images using a Dual Residual Blocks pre-trained model.
• Employing an attention-based mechanism and ensemble feature selection to prioritize relevant features.
• Integrating feature fusion from both normal and segmented MRI images to enhance classification performance.

Main Results:

• Achieved high classification accuracy across multiple datasets: 99.62% (Figshare), 99.54% (Kaggle), 99.39% (Sartaj), and 99.57% (combined).
• Demonstrated significant improvements in accuracy, precision, recall, and F1-score compared to existing models.
• Validated the effectiveness of FusionNet on diverse brain tumor datasets.

Conclusions:

• FusionNet offers a robust and efficient tool for brain tumor classification, significantly improving diagnostic outcomes.
• The study contributes a valuable method to the scientific community for advancing medical diagnosis.
• The proposed model supports medical professionals in achieving superior diagnostic results for brain tumors.