Enhancing Ultrasound Image Quality Across Disease Domains: Application of Cycle-Consistent Generative Adversarial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Processing

Background:

• Existing ultrasound image enhancement methods often use registered images or organ-specific features, limiting generalizability.
• Portable ultrasound devices produce lower quality images compared to high-end equipment, creating a performance gap.
• A universal framework for improving ultrasound image quality across devices and conditions, independent of registration or organ characteristics, is needed.

Purpose of the Study:

• To develop a robust framework for enhancing ultrasound image quality, especially from low-cost portable devices.
• To process nonregistered ultrasound image pairs effectively across various clinical settings and device types.
• To provide a generalized and adaptable solution for improving diagnostic imaging accessibility and quality.

Main Methods:

• A retrospective analysis using a CycleGAN framework enhanced with perceptual loss was employed.
• Perceptual loss was integrated to preserve anatomical integrity by comparing deep features from pretrained neural networks.
• Model performance was evaluated against high-resolution images and validated on a diverse public dataset.

Main Results:

• The enhanced CycleGAN framework significantly outperformed the stable CycleGAN in key evaluation metrics.
• Structural similarity index improved to 0.2889 (vs. 0.2502), peak signal-to-noise ratio to 15.8935 (vs. 14.9430).
• Learned perceptual image patch similarity score was 0.4490 (vs. 0.5005), demonstrating superior enhancement and anatomical detail preservation.

Conclusions:

• The enhanced CycleGAN model successfully bridges the quality gap in ultrasound imaging between different devices.
• Processing nonregistered image pairs, the model enhances visual quality while preserving crucial anatomical structures for diagnosis.
• This approach can democratize high-quality ultrasound imaging, improving healthcare outcomes, especially in resource-limited settings.