Machine learning using supervised approaches has been widely applied to ultrasound image analysis. In contrast, reinforcement learning (RL), which is well suited for sequential decision-making, is underexplored. Ultrasound workflows involve sequential subtasks like image acquisition, quality assessment, summarization, and interpretation that can be integrated into RL frameworks. This scoping review examined RL applications in ultrasound. A comprehensive search was conducted in Scopus, PubMed, Embase, and MEDLINE for studies published between 2015 and 2026. Eligible studies used RL with clinical ultrasound data, and the data were summarized by application area, methods, and anatomical targets. The review also provides a brief overview of RL concepts as foundational knowledge for understanding various RL formulations used. From 326 records retrieved, 39 studies were included. Most studies used model-free deep RL, with value-based methods being the most common, particularly Deep Q-Network (DQN) and its variants, on retrospective data across diverse anatomical targets, with breast, fetal, and uterine ultrasound being the most frequently represented categories. RL was used to automate tasks including navigation, plane localization, landmark detection, and video summarization. RL in ultrasound imaging is an emerging field of research and has advantages for sequential workflow optimization tasks. Most approaches are at an early stage and have been tested on small datasets, lack consistent evaluation protocols, and report limited clinical translation.

(1) Background: Ultrasound (US) imaging is widely used in clinical diagnosis but is often degraded by speckle noise, which reduces image quality and can hinder interpretation. Deep learning (DL) has emerged as a promising approach for US denoising, yet its clinical applicability remains unclear. (2) Methods: A scoping review of studies published in the last four years on DL-based US denoising was conducted following PRISMA-ScR guidelines. Searches were performed in IEEE-Xplore, PubMed, ScienceDirect, Scopus, Web of Science, and Google Scholar. Data was extracted on anatomy, noise type, learning paradigm, network architecture, datasets, evaluation metrics, and performance outcomes. (3) Results: From 951 records retrieved, 36 studies were included. Most focused on breast, fetal, cardiac, and abdominal US. Convolutional neural networks (CNNs), particularly U-Net, were the most common approach, while generative adversarial network, vision transformers, and variational autoencoders were less explored. Reported peak signal-to-noise ratio ranged from 30 to 45 dB and structural similarity index measure from 0.85 to 0.97. Most studies (34 out of 36) relied on synthetic noise, 2D images and paired datasets, with limited evaluation on real clinical images. (4) Conclusion: Supervised CNN-based methods dominate US image denoising, but clinical translation is limited by reliance on synthetic data. Non-paired and no-ground-truth learning approaches remain underexplored despite their suitability for US imaging. Progress is further hindered by inconsistent evaluation protocols, limited robustness assessment on clinical tasks, and restricted dataset access. Future work should focus on standardized clinically meaningful evaluation, openly available datasets, and clinical validation to improve reliability and generalizability of DL-based US denoising methods.

Point-of-care ultrasound (POCUS) devices are increasingly used in pediatric emergency departments for musculoskeletal (MSK) imaging, as they are highly portable and cost-effective. Most POCUS systems use delay-and-sum beamforming (DASB) due to its low computational complexity. The image quality produced by DASB is not optimized for MSK use cases, resulting in images that are often inadequate for reporting purposes. More advanced adaptive beamforming algorithms, such as minimum variance distortionless response beamforming (MVDRB), leverage spatial data correlation to improve resolution and contrast. However, MVDRB is computationally intensive and requires access to raw radio-frequency (RF) data, which is typically not available in commercial POCUS devices. We propose POCUS-Net, a computationally efficient deep learning (DL) based enhancement framework that transforms DASB images to MVDRB-like image quality without requiring raw RF data. The model was built on a PatchGAN-based architecture and incorporates an integrated backscatter (IBS) derived intensity map as a conditioning input, functioning as an attention mechanism to focus on clinically relevant structures. POCUS-Net was trained on in-vivo MSK datasets with input as single-plane wave (SPW) DASB data and output as multiple-plane wave (MPW) MVDRB data. The model was tested on pediatric clinical data sets acquired from three different POCUS devices: Clarius, Philips Lumify, and Telemed. The performance of POCUS-Net was evaluated using contrast ratio (CR), signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and edge preservation index (EPI), a region-based metric specifically developed to assess structural fidelity in MSK bone-enhanced ultrasound images. Compared to baseline DASB images, the POCUS-Net improved CR by up to 40%, SNR by over 200%, and CNR by nearly 130% across the different probes. POCUS-Net images showed substantial improvements in image contrast and bone boundary clarity. The model outperformed common DL models, including CNN-based and GAN-based models, achieving +8.5% in CR, +47.8% in SNR, +6.1% in CNR, and +0.03% in EPI, surpassing the second-best model. We also performed second-order texture analysis using the gray-level run length matrix (GLRLM) to quantify bone structural patterns in both normal and fractured cases, and the gray-level co-occurrence matrix (GLCM) features to assess texture similarity. The inference time of the POCUS-Net was 4.0 ms/image, compared to 148 seconds for conventional MVDRB. The model is fast and requires 219G FLOPs for inference, which is suitable for POCUS.

(1) Background: Ultrasound (US) imaging is widely used in clinical diagnosis but is often degraded by speckle noise, which reduces image quality and can hinder interpretation. Deep learning has emerged as a promising approach for US denoising, yet its clinical applicability remains unclear. (2) Methods: A systematic review of studies published in the last three years on deep learning-based US denoising was conducted following PRISMA-DTA guidelines. Searches were performed in IEEE-Xplore, PubMed, ScienceDirect, Scopus, Web of Science, and Google Scholar. Data were extracted on Anatomy, noise type, learning paradigm, network architecture, datasets, evaluation metrics, and performance outcomes. (3) Results: from 951 records scrapped, 36 studies were included. Most focused-on breast, fetal, cardiac, and abdominal US. Convolutional neural networks (CNNs), particularly U-Net, were the most common approach, while GANs, transformers, and variational autoencoders were less explored. Reported PSNR ranged from 30-45 dB and SSIM from 0.85-0.97. Most studies (34 out of 36) relied on synthetic noise and paired datasets, with limited evaluation on real clinical images. (4) Conclusions: CNN-based methods dominate US denoising research, but translation to clinical practice is limited due to reliance on synthetic data and inconsistent evaluation metrics. Future work should focus on large benchmark datasets and standardized metrics to improve generalizability across clinical settings.