In conventional ultrasound (US) imaging, it is common to transmit several focused beams at multiple locations to generate a multi-focus image with constant lateral resolution throughout the image. However, this method comes at the expense of a loss in temporal resolution, which is important in applications requiring both high-frame rate and constant lateral resolution. Moreover, relative motions of the target with respect to the probe often exist due to hand tremors or biological motions, causing blurring artifacts in the multi-focus image. This article introduces a novel approach for multi-focus US image recovery based on Generative Adversarial Network (GAN) without a reduction in the frame-rate. Herein, a mapping function between the single-focus US image and multi-focus version for having a constant lateral resolution everywhere is estimated through different GANs. We use adversarial loss functions in addition to Mean Square Error (MSE) to generate more realistic ultrasound images. Moreover, we use the boundary seeking method for improving the stability of training, which is currently the main challenge in using GANs. Experiments on simulated and real phantoms as well as on ex vivo data are performed. Results confirm that having both adversarial loss function and boundary seeking training provides better results in terms of the mean opinion score test. Furthermore, the proposed method enhances the resolution and contrast indexes without sacrificing the frame-rate. As for the comparison with other approaches which are not based on NNs, the proposed approach gives similar results while requiring neither channel data nor computationally expensive algorithms.

中文翻译：

---

使用生成对抗网络的快速多焦点超声图像恢复

在传统的超声 (US) 成像中，通常会在多个位置发射多个聚焦波束，以生成在整个图像中具有恒定横向分辨率的多焦点图像。然而，这种方法的代价是时间分辨率的损失，这在需要高帧率和恒定横向分辨率的应用中很重要。此外，由于手部颤抖或生物运动，目标相对于探头的相对运动经常存在，从而导致多焦点图像中的模糊伪影。本文介绍了一种基于生成对抗网络 (GAN) 的多焦点美国图像恢复的新方法，而不会降低帧速率。在此处，通过不同的 GAN 估计单焦点 US 图像和多焦点版本之间的映射函数，以便在各处具有恒定的横向分辨率。除了均方误差 (MSE) 之外，我们还使用对抗性损失函数来生成更逼真的超声图像。此外，我们使用边界寻找方法来提高训练的稳定性，这是目前使用 GAN 的主要挑战。对模拟和真实体模以及离体数据进行了实验。结果证实，在平均意见分数测试方面，同时具有对抗性损失函数和边界寻求训练提供了更好的结果。此外，所提出的方法在不牺牲帧速率的情况下提高了分辨率和对比度指数。至于与其他非基于神经网络的方法的比较，