Cross-Parametric Generative Adversarial Network-Based Magnetic Resonance Image Feature Synthesis for Breast Lesion...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is vital for breast cancer diagnosis, offering insights into tumor morphology and physiology.
• DCE-MRI requires contrast agents and longer acquisition times compared to T2-weighted imaging (T2WI).
• Synthesizing images across different MRI sequences remains a significant challenge in medical imaging.

Purpose of the Study:

• To develop a novel cross-parametric generative adversarial network (GAN)-based feature synthesis (CPGANFS) method.
• To generate discriminative DCE-MRI features from T2WI for enhanced breast cancer diagnosis.
• To provide a framework for generating cross-parametric MR image features from single-sequence images.

Main Methods:

• Proposed a cross-parametric generative adversarial network (GAN)-based feature synthesis (CPGANFS) approach.
• Utilized a Wasserstein GAN with gradient penalty to differentiate generated features from ground-truth DCE-MRI features.
• Decoded T2W images into latent cross-parameter features to reconstruct both DCE-MRI and T2WI features.

Main Results:

• The synthesized DCE-MRI feature-based model achieved a higher prediction performance (AUC = 0.866) compared to the T2WI-based model (AUC = 0.815) (p = 0.036).
• CPGANFS demonstrated improved diagnostic accuracy in breast cancer detection.
• Model visualization indicated enhanced attention to lesion and surrounding parenchyma areas due to learned interparametric information.

Conclusions:

• The CPGANFS method effectively generates DCE-MRI features from T2WI, improving breast cancer diagnostic performance.
• This approach offers a valuable framework for cross-parametric MR image feature generation, enhancing interpretability and predictive power.
• The study highlights the potential of AI-driven feature synthesis to overcome limitations of traditional MRI acquisition protocols.