Advancements in Real-Time Ultrasound Imaging Enhancement Using Deep Learning: A Review

Abstract

This review investigates the impact of the use of deep learning for the enhancement of real-time ultrasound imaging, a vital diagnostic technique in healthcare. Despite their widespread availability, traditional ultrasound machines frequently suffer from speckle noise, low resolution, operator dependence, and suboptimal image quality, all of which lower the confidence and reliability of providing diagnoses from abdominal images. With recent advancements, there have been noteworthy attempts to use deep learning techniques such as convolutional neural networks (CNNs), generative adversarial networks (GANs), U-Net, and encoder–decoder networks for image enhancement, denoising, segmentation, and interpretation. While U-Net and encoder–decoder models have become popular for biomedical imaging for feature extraction and noise removal, GANs are known to generate images in a realistic fashion while maintaining anatomical details. Attention mechanisms and reinforcement learning may further improve real-time performance and reduce computational cost. However, clinical implementation remains challenging because of limited high-quality annotated datasets, computational requirements, development complexity, privacy concerns, and regulatory constraints. This review summarizes the current state of technical and clinical problems in applying deep learning to ultrasound imaging while stressing the need to coordinate researchers, physicians, and regulators to achieve safe and reliable clinical applications