Evaluation of medical image enhancement remains challenging, as conventional full-reference metrics often correlate poorly with perceptual quality and clinical usability. We propose CRISP, a clinically relevant and interpretable scoring protocol that uses vision-language models (VLMs) to approximate radiologist-style assessment in a scalable and reproducible manner. CRISP formulates evaluation as an input-conditioned triplet comparison, jointly considering the degraded input, enhanced output, and reference image. In addition to an overall acceptability score, CRISP provides interpretable attribute-level feedback reflecting clinically relevant cues. Reliability is further improved through evaluator qualification tests that filter inconsistent VLM evaluators. Experiments on low-field to high-field brain MRI enhancement demonstrate that CRISP provides interpretable VLM-based judgments that show consistent agreement trends with radiologist preferences, with the leading VLM evaluator achieving up to 77% Top-1 agreement with expert judgments, while remaining modality-agnostic.

Low-field (LF) brain Magnetic Resonance Imaging (MRI) suffers from degraded image quality that limits its clinical usability. Existing LF to high-field (HF) enhancement methods are mainly optimized for uniform reconstruction fidelity, which often oversmooths subtle and localized abnormalities. We propose a report-aware latent diffusion framework for LF brain MRI enhancement that incorporates high-level clinical semantics into the enhancement process. Diagnostic reports are used during training to learn a shared vision-language embedding space via CLIP-style alignment. The learned semantic representations are injected into a latent diffusion model through cross-attention to modulate the denoising trajectory in a context-dependent manner. This enables selective enhancement of diagnostically relevant structures while preserving global anatomical consistency. The proposed method operates solely on LF MRI at inference time and requires no reports or annotations. Experiments demonstrate improved perceptual quality and structural preservation over image-only diffusion baselines.

Polarization-sensitive optical coherence tomography (PS-OCT) is a powerful imaging modality that captures both structural and polarization-related tissue features, offering significant diagnostic value. Among these, the degree of polarization uniformity (DOPU) is critical for characterizing tissue microstructure. However, obtaining DOPU images typically requires specialized hardware and complex system configurations. To address this limitation, we propose a knowledge-guided deep generative framework, signal attention GAN (SA-GAN), to synthesize DOPU images directly from standard OCT intensity scans. SA-GAN integrates a signal-guided attention mechanism inspired by signal-to-noise ratio (SNR) principles, enabling selective focus on regions with meaningful polarization patterns while suppressing noise-dominated areas. This design allows for the generation of accurate, high-fidelity DOPU images without additional imaging hardware. We validated SA-GAN on three independent datasets: SKIN-PSOCT, CARTILAGE-PSOCT, and the public Retinal-OCT2017 dataset. On SKIN-PSOCT, SA-GAN achieved a structural similarity index measure (SSIM) of 97.8% and a peak signal-to-noise ratio (PSNR) of 28.6 dB. On CARTILAGE-PSOCT, it reached an SSIM of 93.9% and a PSNR of 24.8 dB in cross-dataset testing. Applied to the Retinal-OCT2017 dataset, SA-GAN achieved state-of-the-art performance in a four-class retinal disease classification task. These results demonstrate the robustness and generalizability of our method. SA-GAN provides a cost-effective and practical solution to extend PS-OCT capabilities, supporting the development of intelligent imaging systems for biomedical diagnostics and digital medicine. Our code will be available via this link: https://github.com/Yuhengw/SA-GAN.

Federated learning (FL) enables collaborative model training across decentralized medical datasets while preserving data privacy. Its practical adoption remains limited due to data heterogeneity, specifically, differences in input imaging modality (e.g., CT or MRI) and client task (e.g., segmentation or classification) across participating institutions (clients). Such data heterogeneity poses significant challenges for jointly learning a unified global model that generalizes across clients with different input modality and task. To address this, we propose FedCMT, a modality-agnostic FL framework that adaptively aggregates heterogeneous client models. FedCMT supports flexible input modalities and diverse local tasks by incorporating group-wise adapters and personalized decoders that capture modality- and task-specific features. To enhance collaboration across clients, FedCMT employs a conflict-averse module that extracts modality-invariant representations and mitigates inter-client feature conflicts. FedCMT also integrates a global-to-local knowledge distillation mechanism to balance global consistency and local specialization. The proposed FedCMT maintains stability while fostering shared knowledge in diverse medical imaging modalities. We evaluate FedCMT on ten CT and MR datasets involving up to eight federated clients performing segmentation or classification tasks. Experimental results show that FedCMT consistently outperforms state-of-the-art FL baselines, yielding an average improvement of 4.76% over state-of-the-art methods and 4.01% over standalone training. These results demonstrate FedCMT as a promising adaptable FL for real-world medical image analysis.

With the continuous development of artificial intelligence, deep learning techniques have shown strong performance in many multimedia forensic tasks, especially in image steganalysis. Nevertheless, most current deep learning-based steganalysis approaches are still formulated as binary classification problems, which aim to determine only whether an image contains hidden information. Although such methods achieve strong detection capability, they cannot estimate how much information is embedded in a stego image. To address this limitation, this paper introduces a quantitative deep learning-based steganalysis framework for embedding-rate estimation, where the network directly predicts the embedding rate of an input image. In the proposed framework, embedding rates are used as soft-label supervision signals. The framework is optimized using a Soft-label Cross Entropy (SCE) loss together with two independent L2 loss terms for cover and stego samples, respectively. Experimental results on the BOSSBase and BOWS datasets demonstrate that the proposed framework can effectively estimate embedding rates and achieve reliable quantitative steganalysis performance.

Image super-resolution (SR) aims to reconstruct high resolution images with both high perceptual quality and low distortion, but is fundamentally limited by the perception-distortion trade-off. GAN-based SR methods reduce distortion but still struggle with realistic fine-grained textures, whereas diffusion-based approaches synthesize rich details but often deviate from the input, hallucinating structures and degrading fidelity. This tension raises a key challenge: how to exploit the powerful generative priors of diffusion models without sacrificing fidelity. To address this, we propose SpaSemSR, a spatial-semantic guided diffusion framework with two complementary guidances. First, spatial-grounded textual guidance integrates object-level spatial cues with semantic prompts, aligning textual and visual structures to reduce distortion. Second, semantic-enhanced visual guidance with a multi-encoder design and semantic degradation constraints unifies multimodal semantic priors, improving perceptual realism under severe degradations. These complementary guidances are adaptively fused into the diffusion process via spatial-semantic attention, suppressing distortion and hallucination while retaining the strengths of diffusion models. Extensive experiments on multiple benchmarks show that SpaSemSR achieves a superior perception-distortion balance, producing both realistic and faithful restorations.

Adverse weather conditions, low-light environments, and bumpy road surfaces pose significant challenges to SLAM in robotic navigation and autonomous driving. Existing datasets in this field predominantly rely on single sensors or combinations of LiDAR, cameras, and IMUs. However, 4D millimeter-wave radar demonstrates robustness in adverse weather, infrared cameras excel in capturing details under low-light conditions, and depth images provide richer spatial information. Multi-sensor fusion methods also show potential for better adaptation to bumpy roads. Despite some SLAM studies incorporating these sensors and conditions, there remains a lack of comprehensive datasets addressing low-light environments and bumpy road conditions, or featuring a sufficiently diverse range of sensor data. In this study, we introduce a multi-sensor dataset covering challenging scenarios such as snowy weather, rainy weather, nighttime conditions, speed bumps, and rough terrains. The dataset includes rarely utilized sensors for extreme conditions, such as 4D millimeter-wave radar, infrared cameras, and depth cameras, alongside 3D LiDAR, RGB cameras, GPS, and IMU. It supports both autonomous driving and ground robot applications and provides reliable GPS/INS ground truth data, covering structured and semi-structured terrains. We evaluated various SLAM algorithms using this dataset, including RGB images, infrared images, depth images, LiDAR, and 4D millimeter-wave radar. The dataset spans a total of 18.5 km, 69 minutes, and approximately 660 GB, offering a valuable resource for advancing SLAM research under complex and extreme conditions. Our dataset is available at https://gongweisheng.github.io/DIDLM.github.io/

Automated chest X-ray report generation has recently benefited from reinforcement learning (RL) and large language models. However, RL training often suffers from instability or limited exploration due to fixed Kullback-Leibler (KL) regularization and a static reference policy that accumulates KL pressure over time. We propose Response-Weighted and Validation-Anchored Policy Optimization (REVA-PO), a RL framework that stabilizes long-term training via Response-Weighted Regularization (RER) and Validation-Anchored Policy Reset (VAPR). RER dynamically adjusts per-response KL weights based on advantage and reference-policy entropy, relaxing constraints for high-quality responses while tightening them for low-quality ones. Complementarily, VAPR periodically synchronizes the reference and current policies to the best validation checkpoint, resetting accumulated regularization pressure to expand the viable exploration space. To ensure a robust starting point, we employ a three-stage pipeline consisting of warm-up training, classifier-guided supervised fine-tuning, and RL. Extensive evaluations on MIMIC-CXR and IU-Xray demonstrate that REVA-PO sets new state-of-the-art benchmarks in both linguistic quality and clinical accuracy. Notably, BLEU-4 improves by 5.1

The reversal curse describes a failure of autoregressive language models to retrieve a fact in reverse order (e.g., training on “A > B” but failing on “B < A”). Recent work shows that objectives with bidirectional supervision (e.g., bidirectional attention or masking-based reconstruction for decoder-only models) can mitigate the reversal curse. We extend this evaluation to include a vanilla masked language modeling (MLM) objective and compare it to decoder-only masking-based training across four reversal benchmarks and then provide a minimal mechanistic study of how these objectives succeed. We show that reversal accuracy requires training signal that explicitly makes the source entity a prediction target, and we find little evidence that success corresponds to a single direction-agnostic representation of a fact. Instead, representation distances and linear probes are consistent with storing forward and reverse directions as distinct entries, with different indexing geometry for MLM versus decoder-only masking-based training. Our results caution that objective-level “fixes” can improve reversal behavior without necessarily inducing the kind of latent generalization one might expect from a unified concept.

As multimodal large language models (MLLMs) are increasingly deployed across diverse real-world scenarios, their susceptibility to hallucinations - outputs that are inconsistent with the input or factually incorrect - has emerged as a critical bottleneck to broader adoption. To address this, we introduce a novel plug-and-play hallucination mitigation method that operates in a fully non-intrusive manner, requiring no modification to the model architecture or training pipeline. Our approach leverages the geometric properties of local intrinsic dimensionality (LID) to pre-process inputs in the embedding space, selectively optimizing them to reduce their hallucination potential. By aligning the input's embedding structure with regions of lower hallucination likelihood, our method acts as a lightweight yet effective front-end purification module. Experimental results across mainstream MLLMs demonstrate consistent reductions in hallucination rates, suggesting that the proposed method offers an effective, scalable, and model-agnostic solution toward more reliable multimodal understanding.

Steganography enables covert communication, yet DNN-based methods suffer from the need to transmit separate encoders and decoders. To address this issue, we propose DINN, a model steganography framework that integrates Dynamic Sparse Selection (DSS) and Key Activation into Invertible Neural Networks (INNs). By formulating model steganography as a sparse weight-filling problem, we design DSS that employs learnable thresholds to identify carrier-task-irrelevant filters and reserves sparse positions for secret weights that are activated exclusively by an authorized key. As the first attempt to apply model steganography to INNs, DINN unifies secret embedding and extraction within a single architecture, thereby eliminating the need for separate network transmission. Experiments show DINN achieves effective image denoising (carrier task) and image steganography (secret task), while keeping the secret network invisible to unauthorized users.

Medical image retrieval aims to identify clinically relevant lesion cases to support diagnostic decision making, education, and quality control. In practice, retrieval queries often combine a reference lesion image with textual descriptors such as dermoscopic features. We study composed vision-language retrieval for skin cancer, where each query consists of an image to text pair and the database contains biopsy-confirmed, multi-class disease cases. We propose a transformer based framework that learns hierarchical composed query representations and performs joint global-local alignment between queries and candidate images. Local alignment aggregates discriminative regions via multiple spatial attention masks, while global alignment provides holistic semantic supervision. The final similarity is computed through a convex, domain-informed weighting that emphasizes clinically salient local evidence while preserving global consistency. Experiments on the public Derm7pt dataset demonstrate consistent improvements over state-of-the-art methods. The proposed framework enables efficient access to relevant medical records and supports practical clinical deployment.

Conditional Generative Modeling (CCGM) estimates high-dimensional data distributions, such as images, conditioned on scalar continuous variables (aka regression labels). While Continuous Conditional Generative Adversarial Networks (CcGANs) were designed for this task, their instability during adversarial learning often leads to suboptimal results. Conditional Diffusion Models (CDMs) offer a promising alternative, generating more realistic images, but their diffusion processes, label conditioning, and model fitting procedures are either not optimized for or incompatible with CCGM, making it difficult to integrate CcGANs' vicinal approach. To address these issues, we introduce Continuous Conditional Diffusion Models (CCDMs), the first CDM specifically tailored for CCGM. CCDMs address existing limitations with specially designed conditional diffusion processes, a novel hard vicinal image denoising loss, a customized label embedding method, and efficient conditional sampling procedures. Through comprehensive experiments on four datasets with resolutions ranging from 64 & times; 64 to 192 & times; 192, we demonstrate that CCDMs outperform state-of-the-art CCGM models, establishing a new benchmark. Ablation studies further validate the model design and implementation, highlighting that some widely used CDM implementations are ineffective for the CCGM task.

Large Language Models (LLMs) often experience compounding errors during long text generation. Early mistakes can propagate and lead to drift, faulty reasoning, or repetition. While scaling up models improves capabilities, it requires substantial computational resources, and the resulting self-correction behaviour remains unpredictable at inference time. Self-correction is a promising technique for addressing this issue. However, existing approaches have limitations. Supervised training methods can build self-correcting behaviours into models, but require training data collection and lack cross-domain generalizability. Current post-hoc iterative refinement methods operate only at inference time, but must wait for substantial portions of the draft to be generated before providing feedback. This feedback does not guarantee effective guidance, and the same mistake patterns can still reappear. In this paper, we introduce Once-More, a model-agnostic post-hoc self-correction framework that intervenes during generation. Once-More leverages token-level perplexity and feedback from verifiers to provide continuous guided steering of the generation path through a logit redistribution mechanism. This approach essentially helps accumulate "more correct" steps throughout the generation process. Evaluation on multiple benchmarks demonstrates that Once-More achieves state-of-the-art results compared to other self-correction methods. To our knowledge, Once-More is the first post-hoc method to leverage token perplexity and external feedback to perform continuous guided self-correction.

The seven-point checklist (7PCL) is a widely used diagnostic tool in dermoscopy for identifying malignant melanoma by assigning point values to seven specific attributes. However, the traditional 7PCL is limited to distinguishing between malignant melanoma and melanocytic nevi (MN) and falls short in scenarios where multiple skin diseases with appearances similar to melanoma coexist. To address this limitation, we propose a novel diagnostic framework that integrates a clinical knowledge-based topological graph (CKTG) with a gradient diagnostic strategy featuring a data-driven weighting (GD-DDW) system. The CKTG captures both the internal and external relationships among the 7PCL attributes, while the GD-DDW emulates dermatologists’ diagnostic processes, prioritizing visual observation before making predictions. Additionally, we introduce a multimodal feature extraction approach leveraging a dual-attention mechanism to enhance feature extraction through cross-modal interaction and unimodal collaboration. This method incorporates meta-information to uncover interactions between clinical data and image features, ensuring more accurate and robust predictions. Our approach, evaluated on the EDRA dataset, achieved an average AUC of 88.6%, demonstrating superior performance in melanoma detection and feature prediction. This integrated system provides data-driven benchmarks for clinicians, significantly enhancing the precision of melanoma diagnosis.

Electroencephalography (EEG) is a pivotal tool for exploring brain functions. However, the low amplitude of EEG signals renders them inherently susceptible to contamination from diverse physiological and environmental artifacts, including electromyogram artifacts, electrocardiogram interference, and electrical noise from power lines. These contaminants significantly hinder the analysis and interpretation of EEG data, posing substantial challenges for signal processing. Recently, deep learning paradigms have catalyzed significant progress in EEG denoising, with many studies reporting competitive reconstruction fidelity and artifact suppression in benchmarked settings. Despite this progress, there remains a notable gap in the literature regarding comprehensive reviews of deep learning-based EEG denoising strategies. To bridge this gap, we use the end-to-end denoising pipeline as an analytical framework, examining how data/target construction, input representation, modular architecture, objective design, and evaluation strategies influence model assumptions, the interpretation of model performance, and practical utility. We further discuss selective and multi-task denoising strategies, downstream validation, and model deployment as key issues for translating reconstruction performance into usable EEG applications. Finally, we identify future research directions aimed at developing more reliable, interpretable, and practically useful EEG denoising systems, thereby enhancing the utility of EEG technologies in broader applications.

Gearbox is a core component in mechanical transmission systems, and its accurate fault diagnosis is critical for ensuring operational safety of mechanical equipment. Although intelligent diagnosis methods have achieved promising results, they still face significant challenges. First, gearbox is a complex system with tightly coupled components (e.g., gears and bearings). Most existing methods focus on diagnosing isolated components, failing to effectively decouple and identify multi-component compound faults. Second, in actual engineering, fault data is usually scarce and severely imbalanced compared to normal data, which significantly degrades the performance of conventional data-driven models. To address these challenges, a multi-task dynamic balanced learning network is proposed for compound fault decoupling diagnosis with imbalanced data. Specifically, gearbox compound fault diagnosis is reformulated as a component-level multi-task diagnosis problem. First, a shared backbone network is designed to extract common features from vibration signals, followed by two kernel-adaptive task-specific branch networks that hierarchically disentangle component-level fault features for different tasks. Then, an adaptive balance loss is utilized to reweight training samples and alleviate the class imbalance influence. Finally, a task dynamic balancing strategy is incorporated to dynamically coordinate and optimize multiple diagnostic tasks during training. Extensive experiments are conducted on both real-world engineering and public datasets. The results show that the proposed method still achieves over 98% diagnostic accuracy on both gear and bearing diagnosis tasks even under severe class imbalance condition, demonstrating its superiority and robustness compared to other advanced methods.

Proton Density Fat Fraction (PDFF) is the gold standard for non-invasive fatty liver diagnosis, but its reliance on Magnetic Resonance Imaging (MRI) limits broad clinical applicability. Motivated by the accessibility of B-mode Ultrasound (US) in fatty liver assessment, we propose a novel framework for liver segmentation and PDFF prediction from US images. To enhance generalization ability despite limited paired US-PDFF data, our framework integrates a cross-task self-supervised pretext task that extracts semantic features to guide echo intensity capture, benefiting both liver segmentation and PDFF prediction. To address the noise and artifacts inherent in US images, our framework leverages T2-weighted imaging (T2WI) exclusively during training to establish a feature bridge between US and PDFF, thereby enhancing PDFF prediction. Once trained, the model relies solely on US for inference, making it a practical and cost-effective alternative to MRI-based PDFF estimation. Additionally, our framework introduces an uncertainty-augmented adversarial loss function to refine liver boundary delineation, further improving segmentation and PDFF prediction accuracy. Experimental results demonstrate that our method outperforms state-of-the-art methods in liver segmentation and PDFF prediction; and in a specific application study, our predicted PDFF achieves accuracy comparable to real PDFF for hepatic steatosis classification, highlighting its clinical potential. The full source code and detailed documentation are publicly available at https://github.com/D0ngZhang/SSTB.

Screen-shooting has become a major threat to image watermarking in the era of mobile devices. Most existing approaches focus on increasingly sophisticated distortion simulation during training, yet still adopt short-range network backbones. Such local-scale feature fusion is insufficient to model cross-region dependencies under nonlinear screen-shooting degradations, leading to unreliable decoding. In this paper, we propose WNet, an end-to-end watermarking framework based on a cascaded dual-UNet architecture. WNet establishes long-range dependencies to effectively capture global distortion patterns and enhance resilience against complex real-world degradations. To mitigate the attenuation of fine details introduced by long-range modeling, we design a Laplacian-pyramid-based high-frequency enhancement channel attention module to explicitly strengthen high-frequency structures. In addition, a texture edge guided loss is constructed to align watermark information with invisible texture-rich regions. Experimental results demonstrate that the proposed scheme significantly outperforms existing state-of-the-art methods. On average, our method achieves 1.28% higher bit accuracy and a 0.52 dB gain in terms of PSNR compared with DoBMark.

Despite recent advances in human pose estimation (HPE), poor generalization to out-of-distribution (OOD) data remains a difficult problem. While previous works have proposed Test-Time Adaptation (TTA) to bridge the train-test domain gap by refining network parameters at inference, the absence of ground-truth annotations makes it highly challenging and existing methods typically increase inference times by one or more orders of magnitude. We observe that (1) not every test time sample is OOD, and (2) HPE errors are significantly larger on distal keypoints (wrist, ankle). To this end, we propose ESCAPE: a lightweight correction and selective adaptation framework which applies a fast, forward-pass correction on most data while reserving costly TTA for OOD data. The free energy function is introduced to separate OOD samples from incoming data and a correction network is trained to estimate the errors of pretrained backbone HPE predictions on the distal keypoints. For OOD samples, we propose a novel self-consistency adaptation loss to update the correction network by leveraging the constraining relationship between distal keypoints and proximal keypoints (shoulders, hips), via a second “reverse” network. ESCAPE improves the distal MPJPE of five popular HPE models by up to 7% on unseen data, achieves state-of-the-art results on two popular HPE benchmarks, and is significantly faster than existing adaptation methods.