Generative adversarial network: a statistical-based deep learning paradigm to improve detecting breast cancer in...

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Thermography offers a safe, accessible method for cancer diagnosis.
• Deep learning excels at interpreting complex thermographic images.
• Limited training data hinders deep learning in thermography due to its novelty.

Purpose of the Study:

• To address the challenge of limited training data in deep learning for thermography.
• To improve the accuracy and efficiency of cancer detection using thermograms.

Main Methods:

• Utilized statistical learning and generative adversarial networks (GANs) to estimate data distribution.
• Reconstructed synthetic thermographic images based on estimated statistical distributions.
• Augmented the training dataset with GAN-generated images.

Main Results:

• Achieved significant improvements in distinguishing healthy from cancerous thermograms.
• Enhanced sensitivity by 3-9% and accuracy by 3-7% compared to non-GAN methods.
• Observed up to a 9% increase in specificity, with minor drops in some cases.

Conclusions:

• GAN-based data augmentation effectively overcomes data limitations in thermographic deep learning.
• The proposed method significantly boosts diagnostic performance metrics for cancer detection.
• This approach offers a viable solution for advancing AI-driven medical imaging analysis.