TSGET: Two-Stage Global Enhanced Transformer for Automatic Radiology Report Generation

In clinical diagnosis, radiological reports are essential to guide the patient’s treatment. However, writing radiology reports is a critical and time-consuming task for radiologists. Existing deep learning methods often ignore the interplay between medical findings, which may be a bottleneck limiting the quality of generated radiology reports. Our paper focuses on the automatic generation of medical reports from input chest X-ray images. In this work, we mine the associations between medical discoveries in the given texts and construct a knowledge graph based on the associations between medical discoveries. The patient’s chest X-ray image and clinical history file were used as input to extract the image–text hybrid features. Then, this feature is used as the input of the adjacency matrix of the knowledge graph, and the graph neural network is used to aggregate and transfer the information between each node to generate the situational representation of the disease with prior knowledge. These disease situational representations with prior knowledge are fed into the generator for self-supervised learning to generate radiology reports. We evaluate the performance of the proposed method using metrics from natural language generation and clinical efficacy on two public datasets. Our experiments show that our method outperforms state-of-the-art methods with the help of a knowledge graph constituted by prior knowledge of the patient.

Learning how to capture long-range dependencies and restore spatial information of down-sampled feature maps are the basis of the encoder-decoder structure networks in medical image segmentation. U-Net based methods use feature fusion to alleviate these two problems, but the global feature extraction ability and spatial information recovery ability of U-Net are still insufficient. In this paper, we propose a Global Feature Reconstruction (GFR) module to efficiently capture global context features and a Local Feature Reconstruction (LFR) module to dynamically up-sample features, respectively. For the GFR module, we first extract the global features with category representation from the feature map, then use the different level global features to reconstruct features at each location. The GFR module establishes a connection for each pair of feature elements in the entire space from a global perspective and transfers semantic information from the deep layers to the shallow layers. For the LFR module, we use low-level feature maps to guide the up-sampling process of high-level feature maps. Specifically, we use local neighborhoods to reconstruct features to achieve the transfer of spatial information. Based on the encoder-decoder architecture, we propose a Global and Local Feature Reconstruction Network (GLFRNet), in which the GFR modules are applied as skip connections and the LFR modules constitute the decoder path. The proposed GLFRNet is applied to four different medical image segmentation tasks and achieves state-of-the-art performance.

Chest radiology report generation based on cross-modal multi-scale feature fusion

Camouflaged object detection (COD) aims to accurately detect objects that blend in with their backgrounds. While deep learning-based methods have significantly improved detection accuracy, the persistent similarity in chromatic and textural characteristics between camouflaged objects and their environments continues to pose substantial challenges, particularly in complex scenarios. Existing COD methods still encounter two primary challenges: (1) the difficulty in effectively fusing detailed features with semantic representations, leading to suboptimal balance between local and global features in the fused features; (2) progressive information dilution in conventional top-down decoder architectures, resulting in imprecise boundary localization. To address these dual challenges of multi-scale feature enhancement and semantic information dilution, we propose a D ual-Branch C ross-Layer I nformation F low Network (DCIFNet) for camouflaged object detection in complex scenes. Specifically, our framework integrates a hybrid CNN-ViT architecture for parallel extraction of localized patterns and global contextual features. Subsequently, a global local fusion (GLF) module is employed to iteratively and adaptively fuse the global and local features at each stage using gating weights. Additionally, a novel Dense Connection Decoder (DCD) module is designed to mitigate the information dilution problem associated with traditional top-down decoders. Extensive experiments show that DCIFNet outperforms other mainstream COD models in most performance metrics on four widely used benchmark datasets. On the COD10K dataset, compared with the latest SOTA work of PRNet, DCIFNet improves the metrics of \({S_{\alpha}}\) , \({E_{\phi}}\) , and \(F_{\beta}^{w}\) by 0.11 \(\%\) , 0.11 \(\%\) , and 1.38 \(\%\) , respectively, demonstrating a significant performance enhancement. The code is available https://github.com/WateverOk/DCIFNet .

Dual Graph Convolutional Networks for Chest X-ray Report Generation

Diabetic retinopathy (DR) is frequently discovered in the eyes of diabetics and this has been found to be a contributing factor to vision loss. Early intervention, along with regular fundus photography monitoring, is the most effective way to treat the condition. There are many patients with diabetes who need intensive screening; hence, there is an increase in computer-assisted and fully automated methods for diagnosing DR. Neural networks have come a long way in recent years in various domains. As a result of automating diagnosis of DR and giving customized suggestions to DR patients, it may be seen that precise as well as intricate DR classification is important. Here we have showed cross-modality feature fusion framework for diabetic retinopathy (DR) images categorization. Cross mode here means an RGB image having green channel. A multi-scale and multi-receptive feature extraction block has thus been presented initially so as to learn local and global features from both modalities. Moreover, the learnt features at different scales are successfully fused with current multi-level feature fusion block for image classification task. Our existing system has been compared against state-of-the-art (SOTA) deep learning frameworks for categorizing DR images using the MESSIDOR and IDRID databases. Results analysis shows clearly that the ongoing cross-modality feature fusion based classification framework outperforms all the present SOTA frameworks according to some evaluation metrics.

Precise and rapid ultrasound-based breast cancer diagnosis is essential for effective treatment. However, existing ultrasound image segmentation methods often fail to capture both global contextual features and fine-grained boundary details. This study proposes a dual-branch network architecture that combines the Swin Transformer and Segment Anything Model (SAM) to enhance breast ultrasound image (BUSI) segmentation accuracy and reliability. Our network integrates the global attention mechanism of the Swin Transformer with fine-grained boundary detection from SAM through a multi-stage feature fusion module. We evaluated our method against state-of-the-art methods on two datasets: the Breast Ultrasound Images dataset from Wuhan University (BUSI-WHU), which contains 927 images (560 benign and 367 malignant) with ground truth masks annotated by radiologists, and the public BUSI dataset. Performance was evaluated using mean Intersection-over-Union (mIoU), 95th percentile Hausdorff Distance (HD95) and Dice Similarity coefficients, with statistical significance assessed using two-tailed independent t-tests with Holm-Bonferroni correction ( ). On our proposed dataset, the network achieved a mIoU of 90.82% and a HD95 of 23.50 pixels, demonstrating significant improvements over current state-of-the-art methods with effect sizes for mIoU ranging from 0.38 to 0.61 (p 0.05). On the BUSI dataset, the network achieved a mIoU of 82.83% and a HD95 of 71.13 pixels, demonstrating comparable improvements with effect sizes for mIoU ranging from 0.45 to 0.58 (p 0.05). Our dual-branch network leverages the complementary strengths of Swin Transformer and SAM through a fusion mechanism, demonstrating superior breast ultrasound segmentation performance. Our code is publicly available at https://github.com/Skylanding/DSATNet.

In the harsh underground environment, the transmission gear in the rocker arm of the shearer is susceptible to impact load, frequently leading to failures that result in economic losses or casualties. Therefore, real-time and accurate gear fault diagnosis is crucial. Existing diagnostic models struggle with insufficient feature extraction and low diagnostic accuracy when addressing nonlinear, nonsmooth signals with strong noise in this environment. Therefore, an innovative parallel integration model of Swin Transformer and multiscale convolution is proposed to extract both global and local features under different fault modes and noise conditions. A spatial interaction block (SIB) is introduced into the Swin Transformer branch to enhance global information acquisition. In multiscale convolutional neural network (CNN) branches, depthwise separable convolutions with different expansion coefficients are utilized for multiscale extraction of local features to reduce computational complexity. Considering the differences between global and local features, a feature interaction module (FIM) is added between the two branches to realize the interaction between local and global features, considerably improving the model diagnosis performance. Furthermore, to meet the high real-time diagnosis requirements, a cloud-edge collaboration (CEC) framework is designed to realize the timely alarm of gear faults through the efficient collaboration between edge nodes and the industrial cloud platform. Using the vibration data collected from the rocker arm loading test bed of Taizhong Coal Machinery and the gear data provided by Southeast University for validation, the results show that the diagnostic precision of the proposed method can reach up to 99.55%. Even when the signal-to-noise ratio (SNR) is −2 db, the accuracy is still 93.67%, which is superior to other advanced comparison models, and significantly reduces the completion time of fault diagnosis tasks, and has strong anti-noise and real-time diagnosis capabilities.

Medical Image Semantic Segmentation (MISS), the process of assigning a semantic label to each pixel in an image, is a foundational task in computational medicine, critical for quantitative diagnostics and treatment planning. However, developing robust MISS models faces two intertwined challenges. First, there is an architectural dilemma: Convolutional Neural Networks (CNNs), like U-Net, excel at learning local features but are limited by their receptive fields, failing to capture global context essential for segmenting organs with large deformations. Conversely, Vision Transformers (ViTs) effectively model long-range dependencies but lack the inductive biases of CNNs, leading to poor generalization on the small datasets typical in medicine without extensive pre-training. Second, the prohibitive cost and expertise required for creating pixel-level annotations create a severe data scarcity bottleneck. While Semi-Supervised Learning (SSL) aims to mitigate this by leveraging unlabeled data, existing methods often fail to learn high-level semantic relations and are susceptible to confirmation bias from noisy pseudo-labels, class imbalance, and suboptimal contrastive sample selection. This thesis presents a comprehensive investigation to systematically address these challenges, delivering a cohesive suite of novel deep learning frameworks. The contributions are four-fold: First, to resolve the architectural trade-off, this work introduces CS-Unet, a pure Transformer network built upon a U-Net-like architecture. Its core innovation is the Convolutional Swin Transformer (CST) block, which integrates convolutions directly within the Multi-Head Self-Attention and Feed-Forward Network modules. This design imbues the Transformer with inherent localized spatial context and strong inductive biases, enabling it to efficiently learn both local and global features. Without pre-training, CS-Unet outperforms existing Transformer and CNN-based models on multi-organ and cardiac datasets, achieving state-of-the-art performance with fewer parameters. Second, to address data scarcity, a novel Multi-Scale Cross Supervised Contrastive Learning (MCSC) framework for SSL is developed. MCSC jointly trains CNN and Transformer models, using a cross-teaching paradigm where each network provides pseudo-labels for the other. Crucially, it moves beyond simple output consistency by applying a contrastive loss to feature maps at multiple scales, enforcing hierarchical semantic consistency. To handle the class imbalance endemic to medical imaging, a class-prevalence-aware loss is used to ensure features for infrequent classes are learned robustly. Third, to fortify SSL against noisy pseudo-labels, a certainty-guided contrastive learning strategy is proposed. This approach mitigates the impact of inaccurate pseudo-labels by using a certainty metric to guide the selection of samples for contrastive learning. The framework’s computational efficiency is enhanced through novel sampling strategies that select a few representative samples for contrasting, and a negative memory bank is used to increase sample diversity and eliminate dependence on batch size. Fourth, this thesis introduces a new paradigm for SSL by leveraging external anatomical priors through the Contrastive Cross-Teaching with Registration (CCT-R) framework. CCT-R is the first method to integrate spatial registration transforms into the learning process. It features two novel modules: a Registration Supervision Loss (RSL), which uses transforms between labeled and unlabeled volumes to generate an additional, highly reliable source of pseudo-labels, and Registration-Enhanced Positive Sampling (REPS), which uses registration to identify anatomically-corresponding positive pairs across volumes for contrastive learning. Overall, these contributions provide a powerful toolkit that significantly alleviates the annotation bottleneck in medical AI. The proposed methods demonstrate state-of-the-art performance on challenging segmentation benchmarks, delivering a pathway to develop accurate, data-efficient models for real-world clinical applications and opening new avenues for research into fusing geometric priors with semantic segmentation.

Nocturnal non-speech sounds encapsulate critical physiological and behavioral information, making them a valuable modality for non-invasive assessment of sleep quality. Despite this potential, existing approaches predominantly rely on single-view spectral features or shallow learning architectures, limiting their ability to generalize across diverse acoustic patterns. To overcome these limitations, this study proposes a hybrid deep learning architecture tailored for the classification of seven distinct nocturnal sound categories. The system employs a tri-branch design that independently processes Mel-frequency cepstral coefficients (MFCC), Mel-spectrogram, and constant-Q transform (CQT)-spectrogram representations. Each branch passes through a dedicated pipeline comprising convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM) layers, and attention-equipped vision transformers (ViT). This configuration facilitates hierarchical learning of local, temporal, and global contextual features. The softmax outputs of each branch are fused using a stacking ensemble strategy, with an XGBoost-based meta-classifier performing the final decision integration. A complementary weighted ensemble is also implemented for comparative evaluation. Experimental results on a publicly available seven-class non-speech sound dataset demonstrate the proposed model’s outstanding performance, achieving 99.71% accuracy under 10-fold cross-validation, along with consistently high precision, recall, and F1-scores across all classes. Comparative benchmarks show substantial improvements over existing state-of-the-art models, including CNNs, long short-term memory (LSTM) variants, classical machine learning approaches, and metaheuristic-based ensembles. Supporting analyses such as confidence score distributions and dimensionality reduction visualizations (principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE)) further validate the model’s robustness and discriminative power. These findings highlight the effectiveness of integrating multi-spectral representations, deep hierarchical modeling, and ensemble strategies for high-fidelity nocturnal non-speech sound classification.

Automatic report generation is a technology that automatically generates documents in the form of report by summarizing various materials according to a specific topic in time sequence or subject. Although the main content of the report is text, insertion of appropriate images can improve the completeness of the report. In this paper, we propose an image recommendation method for automatically selecting and inserting appropriate images corresponding to a specific part of a report. In our proposed method, reevaluation of the candidate images is performed based on the semantic similarity between query and the contents of the images. In order to transform semantic information of text query and image into one vector space, we extracted semantic information from image as a set of tags form using deep learning based object detection module. Also, we extracted tags from the given title of the image so that the proposed system can evaluate the candidate images even in the case that the given query includes specific keywords or proper nouns which were not learned by object detection and recognition module in advance. In this paper, we conducted experiments on eight queries related to recent events to verify the applicability of our proposed image recommendation system and evaluate the image selection accuracy.

This paper addresses the problem of automatic generation of natural language descriptions for ontology-described artifacts. The original motivation for the work is the challenge of providing textual narratives of automatically generated scientific workflows (e.g., paragraphs that scientists can include in their publications). The paper presents two systems which generate descriptions of sets of atoms derived from a collection of ontologies. The first system, called nlgPhylogeny, demonstrates the feasibility of the task in the Phylotastic project, providing evolutionary biologists with narrative for automatically generated analysis workflows. nlgPhylogeny utilizes the fact that the Grammatical Framework (GF) is suitable for the natural language generation (NLG) task; the paper shows how elements of the ontologies in Phylotastic, such as web services and information artifacts, can be encoded in GF for the NLG task. The second system, called \(\mathtt{\small nlgOntology}^{A}\), is a generalization of nlgPhylogeny. It eliminates the requirement that a GF needs to be defined and proposes the use of annotated ontologies for NLG. Given a set of annotated ontologies, \(\mathtt{\small nlgOntology}^{A}\) generates a GF suitable for the creation of natural language descriptions of sets of atoms derived from these ontologies. The paper describes the algorithms used in the development of nlgPhylogeny and \(\mathtt{\small nlgOntology}^{A}\) and discusses potential applications of these systems.

Character-level arabic text generation from sign language video using encoder–decoder model

The increasing number of computed tomography (CT) scan examinations and the time-intensive nature of manual analysis necessitate efficient automated methods to assist radiologists in managing their increasing workload. While deep learning approaches primarily classify abnormalities from three-dimensional (3D) CT images, radiologists also incorporate clinical indications and patient demographics, such as age and sex, for diagnosis. This study aims to enhance multilabel abnormality classification and automated report generation by integrating imaging and non-imaging data. We propose a multimodal deep learning model that combines 3D chest CT scans, clinical information reports, patient age, and sex to improve diagnostic accuracy. Our method extracts visual features from 3D volumes using a visual encoder, textual features from clinical indications via a pretrained language model, and demographic features through a lightweight feedforward neural network. These extracted features are projected into a shared representation space, concatenated, and processed by a projection head to predict abnormalities. For the multilabel classification task, incorporating clinical indications and patient demographics into an existing visual encoder, called CT-Net, improves the F1 score to 51.58, representing a increase over CT-Net alone. For the automated report generation task, we extend two existing methods, CT2Rep and CT-AGRG, by integrating clinical indications and demographic data. This integration enhances Clinical Efficacy metrics, yielding an F1 score improvement of for the CT2Rep extension and for the CT-AGRG extension. Our findings suggest that incorporating patient demographics and clinical information into deep learning frameworks can significantly improve automated CT scan analysis. This approach has the potential to enhance radiological workflows and facilitate more comprehensive and accurate abnormality detection in clinical practice.

Stroke is a major factor in causing disability and fatalities. Doctors use computerized tomography (CT) and magnetic resonance imaging (MRI) scans to assess the severity of a stroke. Automatic image segmentation can help doctors diagnose strokes more quickly and accurately, but it is challenging due to the variability of stroke lesions and the limited availability of labeled data. Deep learning is the cutting-edge technique of machine learning and artificial intelligence, which needs an extensive labeled dataset for effective training. Unfortunately, in the medical domain, the availability of labeled data is severely limited, posing a challenge for conventional deep- learning approaches. In this article, we introduce a system that utilizes deep learning in the form of fusing transformer-based and convolutional neural network (CNN)-based features and few-shot learning techniques to segment ischemic strokes in multimedia MRIs. To accomplish this, we employ two different methods. The first method involves parallel fusion, where we combine CNN-based and transformer-based features. The second method utilizes serial fusion, combining CNN-based and transformer models using few-shot learning. Through the integration of transformer and CNN models, we can extract both global and local features and enhance the system's performance. Moreover, we tackle the issue of limited labeled data by integrating few-shot learning techniques. Additionally, our system optimizes efficiency by selecting only the slices with lesions, disregarding unlesioned slices. The system under consideration is trained with the BraTS2020 dataset, evaluated on the ISLES 2015 dataset, and contrasted the performance with cutting-edge systems. The suggested system attains a dice coefficient score of 0.76, surpassing the scores of previous cutting-edge systems by a substantial margin.