With the development of deep learning, the medical image field has also been widely used it to assist in research, and the main research in this paper is to solve the low-dose computed tomography (LDCT) denoising problem using deep learning. Although LDCT reduces the radiation hazard to patients, it also brings more noise which has some visual interference to the doctor's judgment and affects the diagnosis result. To solve this problem, referring to the architecture of cycle-generative adversarial networks (Cycle-GAN) for unsupervised learning, this paper innovatively proposes an end-to-end unsupervised LDCT denoising framework. It combines the U-Net structure for multiscale feature extraction, the attention mechanism for feature fusion, the combination of residual network for feature transformation, and also consider the comparison of GAN network models and introduce perceptual loss to improve the network for the characteristics of medical images. In addition, we build a real LDCT database and design a large number of comparative experiments to validate the method, using both experimental standards in the image field and evaluation standards in the medical field. The main feature of this paper compared with classical methods is that this paper solves the drawback that real data cannot be used for supervised learning, while this experimental result still have quite excellent performance compared with classical excellent methods, which are professionally judged by imaging physicians and meet the clinical needs of physicians.

PET/CT image precise fusion is crucial for integrating functional metabolic information and high-resolution anatomical structures, and can significantly improve the reliability of clinical diagnosis and treatment planning. However, existing GAN-based fusion methods generally have problems such as unstable training, insufficient feature complementarity extraction, and limited cross-modal interaction. To solve the above challenges, this study proposes CMC-GAN, a cross-modal coupled generative adversarial network that combines global-local feature extraction and a dual-discriminator framework. Specifically, the coupled generator with weight sharing achieves bidirectional interaction of PET and CT features, while the global-local feature fusion module can capture fine anatomical details and large-scale contextual information. Two types of modality discriminators jointly constrain the generation process to ensure the complete fidelity of structural and functional in-formation. Experimental results on the Lung-PET-CT-Dx dataset show that CMC-GAN improves the existing best method by an average of 4.46% on six metrics including SCD, MS-SSIM, and so on, verifying its application potential in clinical PET/CT fusion.

With the widespread application of medical imaging in clinical diagnosis and intelligent analysis, the sensitive information contained in such images is facing increasingly severe risks of leakage, making steganography a critical means for safeguarding image privacy. However, existing medical image steganography methods still face significant challenges in the accurate identification and protection of diagnostically relevant regions (ROIs), adaptability to the structural characteristics of medical images, and resistance to advanced steganalysis techniques. To address these issues, this paper proposes MARS-Net, a structure-aware medical image steganography framework that integrates zero-shot medical image segmentation and reinforcement learning-based adaptive probability optimization. The proposed method can precisely localize ROIs without manual annotation, and, through the introduction of an Embeddable Suppression Module (ESM), effectively restricts embedding in structurally high-risk regions such as large all-black backgrounds. In addition, the designed reinforcement learning-based policy network incorporates multiple loss constraints—including ROI exclusion, suppression regions, and modification rate—to achieve dynamic and adaptive optimization of the embedding probability distribution. Extensive experiments on the MSD Pancreas Tumour dataset demonstrate that MARS-Net significantly outperforms traditional steganographic methods in terms of embedding security, resistance to steganalysis, and preservation of image quality in critical regions.

In multi-agent reinforcement learning (MARL), trust region (TR) methods are widely used because they effectively mitigate the nonstationarity of multi-agent systems and facilitate collaboration among diverse agent types. Based on the multi-agent advantage decomposition lemma, TR methods adopt a sequential update scheme (i.e., agents’ policy networks are trained with a certain order). However, current TR methods lack a ranking scheme and train the agents in a random order, this results in suboptimal performance and large variances. To solve this issue, based on agents’ observations (the input of agents’ policy networks), we formulate our ranking criteria and furthermore propose our ranking schemes. Specifically, we avoid agents with similar observations being ranked adjacent to each other for training and give higher priority to the agents with more information in their observations. We extend our schemes to popular TR methods and evaluate them on a series of StarCraftII, Google Football and Multi-Agent MuJoCo tasks, results show that our ranking schemes can enhance current TR methods in many tasks, whatever in performance, efficiency or stability, indicating its modeling capability on both homogeneous and heterogeneous agent tasks.

Image Quality Assessment (IQA) aims to evaluate the perceptual quality of images based on human subjective perception. Existing methods generally combine multiscale features to achieve high performance, but most rely on straightforward linear fusion of these features, which may not adequately capture the impact of distortions on semantic content. To address this, we propose a bottom-up image quality assessment approach based on the Contrastive Language-Image Pre-training (CLIP, a recently proposed model that aligns images and text in a shared feature space), named BPCLIP, which progressively extracts the impact of low-level distortions on high-level semantics. Specifically, we utilize an encoder to extract multiscale features from the input image and introduce a bottom-up multiscale cross attention module designed to capture the relationships between shallow and deep features. In addition, by incorporating 40 image quality adjectives across six distinct dimensions, we enable the pre-trained CLIP text encoder to generate representations of the intrinsic quality of the image, thereby strengthening the connection between image quality perception and human language. Our method achieves superior results on most public Full-Reference (FR) and No-Reference (NR) IQA benchmarks, while demonstrating greater robustness.

Three-dimensional (3D) reconstruction of neurons is a critical and evolving area in neuroscience, addressing the substantial challenges presented by weak signals, high noise levels, and heterogeneous signal distribution in neuronal optical images. Previous methodologies predominantly focused on reconstructing neuronal fibers but faced significant limitations in integrating both neuronal fibers and somas, making it difficult to handle large-scale neuronal image reconstruction. Furthermore, conventional techniques often employ fixed thresholding to eliminate background noise, inadvertently leading to the loss of valuable low-intensity neuronal signals, which are crucial for comprehensive neuronal analysis. In response to these limitations, we propose ADTAH, an innovative neuron image reconstruction method that leverages the adaptive distance transform combined with the Hessian matrix for robust and precise 3D reconstruction of neurons. ADTAH consists of two branches: nerve fiber reconstruction and soma reconstruction. The nerve fiber reconstruction branch begins with adaptive threshold distance transform, dynamically adjusting the Hessian matrix window size to effectively capture nerve fibers of varying thicknesses, thereby optimizing the extraction of tubular structures. The soma reconstruction branch employs high-threshold distance transform to accurately identify and fill somas, ensuring comprehensive coverage of both fibers and somas. Our experiments on publicly available 3D neuron image datasets demonstrate that ADTAH outperforms existing techniques, providing superior differentiation between neuronal and background signals, enhanced computational efficiency, and greater robustness. Our approach has demonstrated state-of-the-art performance on the publicly available 3D Neuron image dataset Big Neuron and an fMost dataset made available by the Allen Institute for Science.