The data processing capability of intelligent reflective surfaces (IRS)-assisted cell-free mobile edge computing (MEC) is constrained by both the offloaded data traffic during wireless transmission and the computed data traffic in the MEC server. The effectiveness of wireless transmission is adversely affected by channel estimation errors. Additionally, fairness is a critical consideration in multi-user scenarios. However, achieving fair multi-user traffic matching in IRS-assisted cell-free MEC networks with imperfect channel state information (CSI) remains an unsolved issue. To tackle this challenge, this paper proposes a max-min multi-user traffic matching (M3TM) strategy for transmission and computing, which mitigates the impact of channel estimation errors through stochastic analysis. The proposed strategy applies a sequential transmission paired with immediate sequential computing mode to ensure IRS gain for each user and improve computing resource utilization efficiency. A cell-free wireless transmission sub-model is proposed with a stochastic analysis of channel estimation errors. Based on the transmission sub-model, a transmission-computing traffic matching model, representing the processed data volume for each user, is proposed, which comprehensively accounts for the mutual constraints between wireless transmission and edge computing. Subsequently, a max-min processed data volume optimization problem is formulated, leveraging the max-min criterion to ensure fairness among users. To address the non-convex formulated problem and alleviate the adverse impact of channel estimation errors, a variable grouping optimization algorithm is proposed, which combines a block coordinate descent (BCD)-based method with a proof-by-contradiction approach to optimize different groups of variables. Simulation results validate the superiority of the proposed strategy.

Fluid antenna-enabled multiple-input multiple-output (FA-MIMO) systems hold significant application potential; however, they also introduce substantial costs in channel state information (CSI) acquisition. In this paper, we propose a novel FA-assisted rectangular differential index modulation (FA-RDIM) scheme for MIMO systems, aiming to achieve high spectral efficiency while addressing the problem of the increased CSI acquisition cost. The transmitted information is mapped to modulation symbols and cyclic shifts of FA pattern indices. A multi-stage detection method is introduced, which reduces computational complexity by identifying the most likely candidates for FA pattern indices. We derive a closed-form expression for the bit error rate (BER) theoretical performance considering error propagation, and present an optimization algorithm based on rank and determinant criterion (RDC), Hamming distance (HD), and gradient descent (GD) to optimize the FA pattern vector set. Simulation results demonstrate that the proposed scheme exhibits a minimal performance loss compared to conventional coherent modulation schemes under static channel conditions, while offering a performance advantage in time-varying channels with outdated CSI.

Gene regulatory relationships can be abstracted as a gene regulatory network (GRN), which plays a key role in characterizing complex cellular processes and pathways. Recently, graph neural networks (GNNs), as a class of deep learning models, have emerged as a useful tool to infer gene regulatory relationships from gene expression data. However, deep learning models have been found to be vulnerable to noise, which greatly hinders the adoption of deep learning in constructing GRNs, because high noise is often unavoidable in the process of gene expression measurement. Can we preferably prototype a robust GNN for constructing GRNs? In this paper, we give a positive answer by proposing a Quadratic Graph Attention Network (Q-GAT) with a dual attention mechanism. We study the changes in the predictive accuracy of Q-GAT and 9 state-of-the-art baselines by introducing different levels of adversarial perturbations. Experiments in the E. coli and S. cerevisiae datasets suggest that Q-GAT outperforms the state-of-the-art models in robustness. Lastly, we dissect why Q-GAT is robust through the signal-to-noise ratio (SNR) and interpretability analyses. The former informs that nonlinear aggregation of quadratic neurons can amplify useful signals and suppress unwanted noise, thereby facilitating robustness, while the latter reveals that Q-GAT can leverage more features in prediction thanks to the dual attention mechanism, which endows Q-GAT with the ability to confront adversarial perturbation. We have shared our code in https://github.com/Minorway/Q-GAT_for_Robust_Construction_of_GRN for readers' evaluation.

The proliferation of wireless terminal devices and services has highlighted the significance of multi-hop and multi-connection wireless sensor networks (MHMC-WSN). However, many challenges have arisen in maintaining effective throughput and end-to-end (E2E) reliability for remote users as the number of hops and connected nodes of relays increases significantly. A cross-layer analysis approach to assess the E2E reliability of MHMC-WSN is proposed in this paper, in which a transmission model based on offloading regions is presented, giving a closed-form solution for both uplink and downlink outage probability in a heterogeneous network. Taking both transmission and scheduling into account, a cross-layer caching queuing model is proposed along with queuing delays analysis for both E2E link and relay nodes. Finally, a hierarchical traffic offloading mechanism (HTOM) is proposed, consisting of a grouping-routing strategy and a dynamic off-load scheduling method. Best-response Stackelberg game and Lyapunov drift-plus-penalty (DPP) optimization theory are utilized to address long-term optimization challenges and slot-by-slot optimization problems in multi-relay scenarios. A 5G-Wi-Fi simulation system for MHMC-WSN is developed to evaluate the E2E transmission reliability and performance of HTOM.

Reconfigurable intelligent surface (RIS)-based index modulation is a promising technology for 6G and beyond networks. In this letter, a novel RIS-assisted reflecting, spatial, and media-based modulation system is proposed with both information decoding and energy harvesting at the receiver. To enhance spectral and energy efficiency, we design an integrated energy management framework which adopts muti-domain index modulation across reflecting, spatial, and media domains. A new energy harvesting-based RIS (EH-RIS) architecture is proposed on the foundation of the multi-domain index modulation, which improves both information transmission and energy harvesting. Compared with traditional RIS architectures, EH-RIS can utilize harvested energy to reinforce signal transmission so as to enhance the transmission reliability. The simulation results show that the proposed scheme can achieve better bit error rate performance than existing index modulation schemes.

Deep learning is a very promising technique for low-dose computed tomography (LDCT) image denoising. However, traditional deep learning methods require paired noisy and clean datasets, which are often difficult to obtain. This paper proposes a new method for performing LDCT image denoising with only LDCT data, which means that normal-dose CT (NDCT) is not needed. We adopt a combination including the self-supervised noise2noise model and the noisy-as-clean strategy. First, we add a second yet similar type of noise to LDCT images multiple times. Note that we use LDCT images based on the noisy-as-clean strategy for corruption instead of NDCT images. Then, the noise2noise model is executed with only the secondary corrupted images for training. We select a modular U-Net structure from several candidates with shared parameters to perform the task, which increases the receptive field without increasing the parameter size. The experimental results obtained on the Mayo LDCT dataset show the effectiveness of the proposed method compared with that of state-of-the-art deep learning methods. The developed code is available at https://github.com/XYuan01/Self-supervised-Noise2Noise-for-LDCT.

Objective. In radiotherapy planning, acquiring both magnetic resonance (MR) and computed tomography (CT) images is crucial for comprehensive evaluation and treatment. However, simultaneous acquisition of MR and CT images is time-consuming, economically expensive, and involves ionizing radiation, which poses health risks to patients. The objective of this study is to generate CT images from radiation-free MR images using a novel quasi-supervised learning framework.Approach. In this work, we propose a quasi-supervised framework to explore the underlying relationship between unpaired MR and CT images. Normalized mutual information (NMI) is employed as a similarity metric to evaluate the correspondence between MR and CT scans. To establish optimal pairings, we compute an NMI matrix across the training set and apply the Hungarian algorithm for global matching. The resulting MR-CT pairs, along with their NMI scores, are treated as prior knowledge and integrated into the training process to guide the MR-to-CT image translation model.Main results. Experimental results indicate that the proposed method significantly outperforms existing unsupervised image synthesis methods in terms of both image quality and consistency of image features during the MR to CT image conversion process. The generated CT images show a higher degree of accuracy and fidelity to the original MR images, ensuring better preservation of anatomical details and structural integrity.Significance. This study proposes a quasi-supervised framework that converts unpaired MR and CT images into structurally consistent pseudo-pairs, providing informative priors to enhance cross-modality image synthesis. This strategy not only improves the accuracy and reliability of MR-CT conversion, but also reduces reliance on costly and scarce paired datasets. The proposed framework offers a practical and scalable solution for real-world medical imaging applications, where paired annotations are often unavailable.

Fast automatic modulation recognition (AMR) is a crucial technique for intelligent communications systems. Short recognition time, consisting of sampling time and calculation time, is the key performance factor of fast AMR. The existing research on fast AMR mostly focuses on reducing the calculation time without considering the impact of sampling time. In this letter, a novel fast AMR method is proposed in which fewer signal samples are collected and utilized to reduce the sampling time. To combat the problem of recognition accuracy loss arising from using fewer signal samples, multimodal deep learning is utilized to guarantee recognition accuracy. Moreover, to solve the problem of model size mismatch between the training and inferring stages, resized signal representation is used to avoid model retraining. Experiments show that the proposed method reduces at least 34.94% of recognition time compared with the traditional single-modality method. In addition, more reduction in recognition time can be achieved at low SNR regions, e.g., up to 72.04% when SNR is 0 dB.

Tuberculosis (TB) and pneumonia remain major global public health challenges, necessitating accurate and efficient diagnostic tools. This study proposes a novel deep learning framework, Large Adaptive Filter and Aligning Normalized Network (LAFAN-Net), designed to improve chest X-ray (CXR) diagnosis by integrating visual and textual information. The framework comprises three key components: (1) a report-guided multi-level alignment mechanism that aligns CXR features with radiology reports at the token, sample, and disease levels; (2) a large adaptive filter block for capturing multi-scale visual patterns; and (3) AlignNorm, a new normalization technique that mitigates oversmoothing and enhances feature separation. LAFAN-Net is evaluated on three publicly available CXR datasets, achieving accuracies of 97.14 %, 95.35 %, and 89.39 %, and F1 scores of 90.77 %, 96.32 %, and 88.33 %, respectively. Extensive ablation studies confirm the model's robustness. The results underscore LAFAN-Net's ability to extract clinically meaningful features while maintaining interpretability, supported by singular value distributions and Gradient-weighted Class Activation Mapping visualizations. Future work will explore extending the model to broader disease categories and multi-class classification tasks to enhance clinical utility. In addition, improving computational efficiency and ensuring real-time applicability are essential for deployment in resource-limited settings.

Alzheimer's Disease (AD) poses significant challenges in neuroimaging, where early and accurate diagnosis is essential for timely intervention. Magnetic Resonance Imaging (MRI) serves as a key modality for AD diagnosis, providing detailed anatomical insights without radiation exposure. While traditional Convolutional Neural Networks (CNNs) excel at local feature extraction from MRI data, they fail to capture global contextual information critical for AD detection. To address this limitation, we propose the Lightweight Robust Alzheimer's Disease Vision Transformer (LRAD-ViT), a computationally efficient framework tailored for early AD detection. LRAD-ViT enhances the global learning capabilities of Vision Transformers through a novel adaptive token fusion technique. The proposed method selectively identifies and merges non-essential tokens within brain MRI scans, optimizing computational efficiency without sacrificing diagnostic performance. By dynamically adapting the token structure during model operation, the method prioritizes diagnostically relevant regions in the images. Additionally, randomized learning regularization improves learning dynamics and model robustness. Additionally, randomized learning regularization improves learning dynamics and model robustness. Validation on two brain MRI datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI) demonstrates that LRADViT achieves a diagnostic accuracy of 93.41 % on AD versus cognitive normal (CN) and 90.95 % on CN versus mild cognitive impairment (MCI), all while reducing computational demands by around 40 %. These metrics substantiate LRAD-ViT's superior diagnostic performance and computational efficiency, making it a promising tool for the early and accurate diagnosis of Alzheimer's Disease, with significant implications for clinical practice.

Automatic modulation classification (AMC) is critical for efficient spectrum management and robust wireless communications. However, AMC remains challenging due to the complex interplay of signal interference and noise. In this work, we propose an innovative framework that integrates traditional signal processing techniques with large-language models (LLMs) to address AMC. Our approach leverages higher-order statistics and cumulant estimation to convert quantitative signal features into structured natural language prompts. By incorporating exemplar contexts into these prompts, our method exploits the LLM’s inherent familiarity with classical signal processing, enabling effective one-shot classification without additional training or preprocessing (e.g., denoising). Experimental evaluations on synthetically generated datasets—spanning both noiseless and noisy conditions—demonstrate that our framework achieves competitive performance across diverse modulation schemes and signal-to-noise ratios (SNRs). Moreover, our approach paves the way for robust foundation models in wireless communications across varying channel conditions, significantly reducing the expense associated with developing channel-specific models. This work lays the foundation for scalable, interpretable, and versatile signal classification systems in next-generation wireless networks.

[1] K. He, X. Chen, S. Xie, Y. Li, P. Doll & aacute;r, and R. Girshick, "Masked autoencoders are scalable vision learners," in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Jun. 2022, pp. 16000-16009. [2] R. Balestriero et al., "A cookbook of self-supervised learning," 2023, arXiv:2304.12210. [3] A. Roberts et al., "Exploring the limits of transfer learning with a unified text-to-text transformer," Google, Tech. Rep., 2019. [4] M. Caron et al., "Emerging properties in self-supervised vision transformers," in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), Oct. 2021, pp. 9650-9660. S. Zheng, S. Chen, T. Chen, Z. Yang, Z. Zhao, and X. Yang, "Deep learning-based SNR estimation," IEEE Open J. Commun. Soc., vol. 5, pp. 4778-4796, 2024. [6] S. Manzoor and N. S. Othman, "Adaptive modulation with CAZAC preamble-based signal-to-noise-ratio estimator in OFDM cooperative communication system," IEEE Access, vol. 10, pp. 126550-126560, 2022. E. Perenda, S. Rajendran, G. Bovet, S. Pollin, and M. Zheleva, "Learning the unknown: Improving modulation classification performance in unseen scenarios," in Proc. IEEE INFOCOM Conf. Comput. Commun., May 2021, pp. 1-10. Y. Shi, H. Xu, Y. Zhang, Z. Qi, and D. Wang, "GAF-MAE: A self-supervised automatic modulation classification method based on gramian angular field and masked autoencoder," IEEE Trans. Cognit. Commun. Netw., vol. 10, no. 1, pp. 94-106, Feb. 2024. [5] [7] [8]

Manual lung ultrasound (LUS) scoring is influenced by clinicians’ subjective interpretation, leading to potential inconsistencies and misdiagnoses due to varying levels of experience. To improve monitoring of pulmonary ventilation and support early diagnosis, we propose an automated LUS scoring network based on an 8-point scale, named the detailed-global fusion residual network (DGF-ResNet). This network combines local and global features using the hybrid feature fusion Block, which includes the detail feature extraction (DFE) and global feature extraction (GFE) Modules. The DFE module employs a local channel and spatial attention mechanism to capture fine details, while the GFE Module utilizes a three-order recursive gated convolution and a global channel and spatial attention mechanism to extract global features. Experimental results on the FCSPF-13324 dataset from the Second Affiliated Hospital of Zhejiang University show that DGF-ResNet outperforms VGG16, ResNet50, and Vision Transformer in accuracy, precision, recall, and F1-score. Specifically, DGF-ResNet improves over Vision Transformer by 7.05, 4.52, and 5.89 percentage points, over VGG16 by 3.06, 4.37, and 3.8 points, and over ResNet50 by 2.05, 4.26, and 3.34 points, respectively.

In the realm of ultra short wave satellite communication channels, various forms of continuous phase signal transmission exist. However, distinguishing between these signals poses a challenge due to their similar frequency domain characteristics. This paper introduces a novel modulation recognition algorithm for continuous phase signals (including MSK, single modulation index CPM, multiple modulation index CPM, SOQPSK, SBPSK) that is based on combined spectral features using ResNeXt-50 with channel and spatial attention mechanism (CS-ResNeXt50). The algorithm first proposes a combined diagrams based on quadratic and quartic spectrum analysis, which exploits the unique characteristics of the impulse spectral lines to differentiate signals with varying continuous phases. Compared to FFT and time-frequency spectrums, the combined diagrams demonstrates superior discrimination capabilities. Simultaneously, the CS-ResNeXt50 network is introduced, incorporating a Channel and Spatial (CS) attention mechanism to enhance feature learning from the combined diagrams. The integration of cross entropy and triplet loss functions further refines feature extraction within the network. Experimental results show that the proposed combined diagrams improves recognition performance by 15.4

Alzheimer's disease (AD) is the most common dementia that is often seen among the elderly. AD can cause the loss of cognitive ability and memory, which can result in death as AD is progressive. The exact cause of AD is still in research, but it is believed to be related to genes, diet, and environment. One observation of AD is the shrinkage of the hippocampus and frontal lobe cortex. Magnetic resonance imaging (MRI) is often employed in the diagnosis of AD as it can produce clear images of the soft tissues. In this study, a new computer-aided diagnosis (CAD) method named RanCom-ViT, is proposed to interpret the brain MRI slices automatically and precisely for AD diagnosis with better global representation learning and efficiency. A pre-trained vision transformer (ViT) is chosen as the backbone because ViTs with attention modules can achieve better performance than convolutional neural networks on larger datasets. Then, a novel token compression block is proposed to improve the efficiency of the RanCom-ViT by removing the less important tokens. Further, the classification head of the RanCom-ViT is enhanced by a random vector functional-link structure to obtain better classification performance in AD diagnosis. A large public brain MRI dataset is utilized in the evaluation experiments of the proposed RanCom-ViT, and it achieved an overall accuracy of 99.54% with a double throughput than the benchmark model. The results reveal that the RanCom-ViT outperforms several existing state-of-the-art AD diagnosis methods in terms of accuracy, and the token compression method contributes to higher efficiency.

IntroductionPulmonary granulomatous nodules (PGN) often exhibit similar CT morphological features to solid lung adenocarcinomas (SLA), making preoperative differentiation challenging. This study aims to address this diagnostic challenge by developing a novel deep learning model.MethodsThis study proposes MAEMC-NET, a model integrating generative (Masked AutoEncoder) and contrastive (Momentum Contrast) self-supervised learning to learn CT image representations of intra- and inter-solitary nodules. A generative self-supervised task of reconstructing masked axial CT patches containing lesions was designed to learn intra- and inter-slice image representations. Contrastive momentum is used to link the encoder in axial-CT-patch path with the momentum encoder in coronal-CT-patch path. A total of 494 patients from two centers were included.ResultsMAEMC-NET achieved an area under curve (95% Confidence Interval) of 0.962 (0.934–0.973). These results not only significantly surpass the joint diagnosis by two experienced chest radiologists (77.3% accuracy) but also outperform the current state-of-the-art methods. The model performs best on medical images with a 50% mask ratio, showing a 1.4% increase in accuracy compared to the optimal 75% mask ratio on natural images.DiscussionThe proposed MAEMC-NET effectively distinguishes between benign and malignant solitary pulmonary nodules and holds significant potential to assist radiologists in improving the diagnostic accuracy of PGN and SLA.

Integrated sensing and communication (ISAC), by combining the communication and sensing functions in shared frequency bands, emerges as a promising technology for future wireless networks. However, the performance of ISAC may be affected by the channel fading and massive connections. In this paper, we propose a non-orthogonal multiple access (NOMA) aided ISAC scheme via intelligent reflective surface (IRS) to set up virtual line-of-sight links for multi-user communication and target sensing. Specifically, our goal is to maximize the sum rate through the joint optimization of active transmit beamforming at the base station and passive phases for reflecting at the IRS, while satisfying the sensing requirement for the target user. Since the original optimization problem is non-convex, we first decompose it into two subproblems, which are converted into convex ones by applying the successive convex approximation. Then, an alternating optimization algorithm is proposed to derive a solution to the original problem. Simulation results validate that the proposed scheme can effectively enhance the multi-user NOMA communication performance while guaranteeing the sensing quality by introducing IRS.

Identifying wireless modulation schemes is essential for cognitive radio, but standard supervised models often degrade under distribution shift, and training domain-specific wireless foundation models from scratch is computationally prohibitive. Large Language Models (LLMs) offer a promising training-free alternative via in-context learning, yet feeding raw floating-point signal statistics into LLMs overwhelms models with numerical noise and exhausts token budgets. We introduce DiSC-AMC, a framework that reformulates Automatic Modulation Classification (AMC) as an LLM reasoning task by combining aggressive feature discretization with nearest-neighbor retrieval over self-supervised embeddings. By mapping continuous features to coarse symbolic tokens, DiSC-AMC aligns abstract signal patterns with LLM reasoning capabilities and reduces prompt length by over 50%. Simultaneously, utilizing a DINOv2 visual encoder to retrieve the k_NN most similar labeled exemplars provides highly relevant, query-specific context rather than generic class averages. On a 10-class benchmark, a fine-tuned 7B-parameter LLM using DiSC-AMC achieves 83.0% in-distribution accuracy (-10 to +10 dB) and 82.50% out-of-distribution (OOD) accuracy (-11 to -15 dB), outperforming supervised baselines. Comprehensive ablations on vanilla LLMs demonstrate the token efficiency of DiSC-AMC. A training-free 7B LLM achieves 71% accuracy using only 0.5 K-token prompt,surpassing a 200B-parameter baseline that relies on a 2.9K-token prompt. Furthermore, similarity-based exemplar retrieval outperforms naive class-average selection by over 20%. Finally, we identify a fundamental limitation of this pipeline. At extreme OOD noise levels (-30 dB), the underlying self-supervised representations collapse, degrading retrieval quality and reducing classification to random chance.

Deep learning has been applied to compressive sensing (CS) of images successfully in recent years. However, existing network-based methods are often trained as the black box, in which the lack of prior knowledge is often the bottleneck for further performance improvement. To overcome this drawback, this paper proposes a novel CS method using non-local prior which combines the interpretability of the traditional optimization methods with the speed of network-based methods, called NL-CS Net. We unroll each phase from iteration of the augmented Lagrangian method solving non-local and sparse regularized optimization problem by a network. NL-CS Net is composed of the up-sampling module and the recovery module. In the up-sampling module, we use learnable up-sampling matrix instead of a predefined one. In the recovery module, patch-wise non-local network is employed to capture long-range feature correspondences. Important parameters involved (e.g. sampling matrix, nonlinear transforms, shrinkage thresholds, step size, $etc.$) are learned end-to-end, rather than hand-crafted. Furthermore, to facilitate practical implementation, orthogonal and binary constraints on the sampling matrix are simultaneously adopted. Extensive experiments on natural images and magnetic resonance imaging (MRI) demonstrate that the proposed method outperforms the state-of-the-art methods while maintaining great interpretability and speed.

Direct sequence spread spectrum (DSSS) communications are highly significant in military and civilian wireless communications because of its ability to resist narrowband interference, multipath interference and high security. However, under low signal-to-noise ratio (SNR), the detection of DSSS signals becomes very difficult under non-cooperative communication. Therefore, in order to improve the detection performance of DSSS signals in fading channels with low SNR, a DSSS signal detection method based on an eigenvalues local binary patterns residual network (EL-ResNet) is proposed with an unknown spread spectrum sequence. Firstly, according to the characteristics of DSSS signals, the eigenvalues of the sample covariance matrix of the DSSS signals are used to construct the signal eigenvalue histograms. This method uses the strong contrast of eigenvalue energy and gradient change to enlarge the discrimination of images with or without DSSS signals. Then, EL-ResNet is built to distinguish between eigenvalue histograms with or without DSSS signals. The local binary patterns (LBPs) are introduced to constrain the loss function of the network, increasing the differentiation of the two types of eigenvalue histograms, and improving the DSSS signal detection performance of the network. Finally, the experimental results show that the performance of the DSSS signal detection method based on EL-ResNet is larger than fast Fourier transform (FFT), FFT-deep neural network (FFT-DNN), covariance matrix-convolutional neural network (CM-CNN) detection methods.