Brain cancer classification based on multistage ensemble generative adversarial network and convolutional neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Accurate brain tumor classification is crucial for effective treatment planning.
• Existing methods face challenges in effectively fusing multimodal data and extracting comprehensive features.
• The integration of deep learning models offers a promising avenue for enhancing diagnostic accuracy.

Purpose of the Study:

• To develop an advanced, high-accuracy model for brain tumor classification.
• To leverage multimodal feature fusion and a dual-path network architecture.
• To improve upon existing methodologies through a novel hybrid approach.

Main Methods:

• Utilized pretrained models (EfficientNet-B7, ResNet-152) and a custom convolutional neural network (CNN) for feature extraction.
• Implemented a two-stage ensemble strategy combining these models.
• Employed Maximum Variance Unfolding (MVU) for relevant attribute selection.
• Integrated a generative adversarial network and Neural Autoregressive Distribution Estimation (NADE-K) within the CNN framework.

Main Results:

• The proposed two-stage ensemble hybrid CNN model achieved a classification accuracy of 99.63%.
• The integration of MVU demonstrated significant improvements in brain tumor classification performance.
• The synergistic combination of multimodal feature fusion and dual-path networks proved highly effective.

Conclusions:

• The developed two-stage ensemble hybrid CNN model represents a significant advancement in brain tumor classification.
• The methodology shows high potential for clinical application in neuro-oncology.
• This approach offers a robust and accurate solution for analyzing complex medical imaging data.