dm-GAN: Distributed multi-latent code inversion enhanced GAN for fast and accurate breast X-ray image automatic...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Breast cancer poses significant health risks, necessitating effective diagnostic tools like mammography.
• AI algorithms enhance mammography by identifying breast masses, but require extensive training data.
• Manual collection and labeling of breast images for AI training are time-consuming and inefficient.

Purpose of the Study:

• To develop a fast, accurate, and automated method for generating synthetic breast images for AI training.
• To address the data scarcity issue in training AI models for breast cancer diagnosis.

Main Methods:

• Proposed a distributed multi-latent code inversion enhanced Generative Adversarial Network (dm-GAN).
• Utilized GAN's generator and discriminator for automatic image synthesis.
• Implemented multi-latent code inverse mapping in the generator to simplify fitting and enhance accuracy.
• Employed a multi-discriminator structure to improve discrimination performance.

Main Results:

• Achieved automatic generation of high-accuracy breast images.
• Demonstrated a 1.84 dB higher Peak Signal-to-Noise Ratio (PSNR) compared to state-of-the-art methods.
• Reported a 5.61% lower Fréchet Inception Distance (FID), indicating superior image quality.
• Generated images 1.38x faster than existing approaches.

Conclusions:

• The proposed dm-GAN effectively generates synthetic breast images with enhanced accuracy and speed.
• This method offers a promising solution for overcoming data limitations in AI-driven breast cancer screening.
• dm-GAN facilitates more efficient development and deployment of AI tools for mammography analysis.