BRMI-Net: Deep Learning Features and Flower Pollination-Controlled Regula Falsi-Based Feature Selection Framework for...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Early breast cancer detection via mammography is crucial for reducing mortality.
• Deep learning excels at feature extraction but can be hindered by redundant data.
• Existing methods require improvement for accuracy and efficiency.

Purpose of the Study:

• To develop a novel framework for breast cancer diagnosis using entropy-controlled deep learning and flower pollination optimization.
• To enhance mammogram image quality and feature extraction for improved diagnostic performance.
• To evaluate the proposed framework's accuracy and efficiency against current methods.

Main Methods:

• A filter fusion-based contrast enhancement technique was applied to mammogram images.
• A pre-trained ResNet-50 model was fine-tuned using transfer learning on original and enhanced datasets.
• Deep features were extracted, combined using serial mid-value features, and optimized with entropy-controlled flower pollination optimization.
• Classifiers including neural networks and machine learning algorithms were employed for diagnosis.

Main Results:

• The proposed framework achieved high accuracy rates of 93.8%, 99.5%, and 99.8% on the CBIS-DDSM, INbreast, and MIAS datasets, respectively.
• Demonstrated significant improvements in accuracy compared to existing methods.
• Showcased a reduction in computational time.

Conclusions:

• The developed framework offers a robust and efficient approach for breast cancer detection using mammograms.
• Entropy-controlled deep learning combined with flower pollination optimization enhances diagnostic accuracy.
• This method holds promise for improving early breast cancer diagnosis and patient outcomes.