Objectives This study investigated alterations in resting-state brain networks in catamenial epilepsy (CE) and their associations with serum sex hormone levels. Methods First, we constructedfunctional networks based on resting-state fMRI datato compute nodal attributes and identify brain regions exhibiting significant group differences. Subsequently, independent component analysis (ICA) identified important networks and characterized their connectivity patterns. Finally, associations between these network metrics and sex hormone levels were examined. Results A total of45 patientswere included in the final analysis, comprising19with CE,26with non-catamenial epilepsy (NCE), and27healthy controls (HC). Nodal efficiency (Ne) differed significantly in key brain regions between patient group and HC group, including the right precentral gyrus (PreCG.R), left and right calcarine cortex (CAL.L and CAL.R), left lingual gyrus (LING.L), right superior parietal gyrus (SPG.R), right inferior parietal lobule (IPL.R) and left precuneus (PCUN.L). While connectivity between the default mode network (DMN) and sensory networks (visual/auditory) was generally weakened in epilepsy patients, CE specifically exhibited a reconfigured attention-network profile: strengthened connectivity between the dorsal attention network (DAN) and auditory network (AN), and weakened connectivity between the ventral attention network (VAN) and visual network (VN). After correction for multiple comparisons, partial correlation analysis controlling for age revealed no statistically significant correlations between sex hormones and brain network metrics. Conclusion CE patients exhibited decreased Ne in critical regions of the AN, DMN and VN, alongside predominant disruptions in DMN connectivity. These alterations may be partially compensated by increased connectivity in the DAN, giving rise to a unique network pathological pattern. The regulatory effects of sex hormones on brain networks require further confirmation in large-scale longitudinal studies.

Pathologic myopia is one of the leading causes of global visual impairment, with myopic maculopathy being its most severe complication. The META-PM classification system provides a standardized framework for grading myopic maculopathy, yet manual assessment remains subjective and time-consuming. This paper proposes Myopic Maculopathy Classification Network(MMC-Net), an innovative deep learning architecture based on ResNet-18, enhanced through the integration of a novel channel attention and feature enhancement modules to achieve automated META-PM classification. Our methodology was evaluated using the Myopic Maculopathy Classification Challenge dataset, comprising 1,391 fundus images distributed across five META-PM categories. The proposed approach achieved remarkable performance, demonstrating superior classification capability and stability compared to state-of-the-art methods and attention mechanisms.

Digital Light Processing (DLP) 3D printing of nanocomposites holds great potential for manufacturing high-performance functional parts. However, balancing printing precision and mechanical properties, such as modulus and toughness, remains a significant challenge due to the complex nonlinear relationships between material composition and printing outcomes. As traditional trial-and-error optimisation is time-consuming and inefficient, here we propose a combined forward-inverse machine learning framework to accelerate the material design process. First, a MLP-based forward model is built to predict performance metrics from material formulas. Subsequently, a novel Gradient Refinement Loop (GRL) is developed for inverse design. By employing a fixed forward ensemble as the performance evaluator, this module iteratively optimises the input formulation through gradient descent to achieve the predefined target performance. The forward surrogate accurately maps the complex design space, achieving high predictive accuracy for printing precision (R & sup2; = 0.94), modulus (R & sup2; = 0.943), and toughness (R & sup2; = 0.90). Empowered by this surrogate, the gradient-refined inverse module identifies precise target formulations with exceptional fidelity, achieving R & sup2; values of 0.95, 0.96, and 0.915 for the respective metrics. The results demonstrate that the proposed data-driven approach effectively identifies optimised recipes that simultaneously satisfy stringent requirements for precision and mechanical strength.

BackgroundOvarian cancer remains one of the deadliest gynecologic malignancies. Poor outcomes largely reflect late diagnosis, marked inter- and intratumoral heterogeneity, and variable treatment response.MethodsThis review summarizes recent advances in artificial intelligence (AI) for ovarian cancer research and clinical care, focusing on imagine-based radiology, digital pathology; longitudinal clinical data/Electronic Health Record (EHR), and spatial-temporal multi-omics.ResultsAI approaches have been applied to tumor detection and classification, prognostic risk stratification, and treatment response prediction. Multimodal models that integrate imaging, molecular profiling, and clinical data enable more refined characterization of tumor heterogeneity and the tumor microenvironment, supporting improved diagnosis, risk assessment, and individualized management.

AIMS:To develop and validate a machine learning-based risk prediction model for delirium in older inpatients. DESIGN:A prospective cohort study. METHODS:A prospective cohort study was conducted. Eighteen clinical features were prospectively collected from electronic medical records during hospitalisation to inform the model. Four machine learning algorithms were employed to develop and validate risk prediction models. The performance of all models in the training and test sets was evaluated using a combination of the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, Brier score, and other metrics before selecting the best model for SHAP interpretation. RESULTS:A total of 973 older inpatient data were utilised for model construction and validation. The AUC of four machine learning models in the training and test sets ranged from 0.869 to 0.992; the accuracy ranged from 0.931 to 0.962; and the sensitivity ranged from 0.564 to 0.997. Compared to other models, the Random Forest model exhibited the best overall performance with an AUC of 0.908 (95% CI, 0.848, 0.968), an accuracy of 0.935, a sensitivity of 0.992, and a Brier score of 0.053. CONCLUSION:The machine learning model we developed and validated for predicting delirium in older inpatients demonstrated excellent predictive performance. This model has the potential to assist healthcare professionals in early diagnosis and support informed clinical decision-making. IMPACT:By identifying patients at risk of delirium early, healthcare professionals can implement preventive measures and timely interventions, potentially reducing the incidence and severity of delirium. The model's ability to support informed clinical decision-making can lead to more personalised and effective care strategies, ultimately benefiting both patients and healthcare providers. REPORTING METHOD:This study was reported in accordance with the TRIPOD statement. PATIENT OR PUBLIC CONTRIBUTION:No patient or public contribution.

Purpose:To evaluate adaptive optics scanning laser ophthalmoscope (AO-SLO) features and identify their multimodal retinal imaging characteristics using AO-SLO, swept-source optical coherence tomography (SS-OCT), and color fundus photography in patients undergoing surgery for epiretinal membrane (ERM), and to provide valuable insights into the mechanism underlying postoperative ERM progression. Methods:Patients with ERM underwent multimodal retinal imaging following surgery, including color fundus photography, SS-OCT, and AO-SLO. AO-SLO features were characterized, and their corresponding multimodal retinal imaging findings were analyzed. Results:Seventeen eyes from 17 postoperative ERM patients were included in this pilot study (mean age [SD], 68.06 [8.26] years; 12 female [70.6%]). Seven categories of AO-SLO features were identified in patients with ERM after surgery. The most common features were granular membrane (n = 17 [100%]), waxy membrane (n = 12 [70.6%]), and microcysts (n = 12 [70.6%]), after dissociated optic nerve fiber layer (n = 11 [64.7%]), microfolds (n = 6 [35.3%]), punctate reflectivity (n = 6 [35.3%]), and retinal folds (n = 5 [29.4%]). Conclusions:AO-SLO imaging provides detailed insights into retinal microstructural changes after surgery for ERM. Further longitudinal studies tracking AO-SLO features over time and integrating multimodal retinal imaging modalities could enhance our understanding of retinal restoration processes and improve the prediction of surgical outcomes in ERM.

Knee medial compartment osteoarthritis is effectively treated by a medial open-wedge high tibial osteotomy (MOWHTO). The feasibility and safety of MOWHTO for mild lateral meniscal tears are unknown. This study examined the feasibility and safety of knee joint weight-bearing line ratio (WBLr) adjustment during MOWHTO with lateral meniscal injuries. We used a healthy adult male's lower extremities computed tomography scans and knee joint magnetic resonance imaging images to create a normal fine element (FE) model. Based on this model, we generated nine FE models for the MOWHTO operation (WBLr: 40–80%) and 15 models for various lateral meniscal injuries. A compressive load of 650N was applied to all cases to calculate the von Mises stress (VMS), and the intact lateral meniscus' maximal VMS at 77.5% WBLr was accepted as the corrective upper limit stress. Our experimental results show that mild lateral meniscal tears can withstand MOWHTO, while severe tears cannot. Our findings expand the use of MOWHTO and provide a theoretical direction for practical decisions in patients with lateral meniscal injuries.

Continuous Intraoperative Neurophysiologic Monitoring (cIONM) is a widely used technology to improve surgical outcomes and prevent cranial nerve injury during skull base surgery. Monitoring of free-running electromyogram (EMG) plays an important role in cIONM, which can be used to identify different discharge patterns, alert the surgeon to potential nerve damage promptly, etc. In this dataset, we collected clinical multichannel EMG signals from 11 independent patients’ data using a Neuromaster G1 MEE-2000 system (Nihon Kohden, Inc., Tokyo, Japan). Through innovative classification methods, these signals were categorized into seven different categories. Remarkably, channel 1 and channel 2 captured continuous EMG signals from the facial nerve (VII cranial nerve), while channel 3 to channel 6 focused on V, XI, X, and XII cranial nerves. This is the first time that intraoperative EMG signals have been collated and presented as a dataset and labelled by professional neurophysiologists. These data can be utilized to develop the architecture of neural networks in deep learning, machine learning, pattern recognition, and other commonly employed biomedical engineering research methods, thereby providing valuable information to enhance the safety and efficacy of surgical procedures.

This article proposes a traffic signal recognition algorithm based on deep learning, addressing prominent issues in traffic signal light detection, such as low detection accuracy, slow speed, and large volume making it difficult to install on mobile devices. The algorithm is built upon the YOLOv5S network module and employs Convolutional Neural Network (CNN) for feature extraction and classification. Two image enhancement techniques, adaptive anchor boxes and adaptive image scaling, are used to improve detection speed and accuracy. Additionally, the Mosaic data augmentation technique is applied to enhance the accuracy and robustness of traffic signal light recognition. Extensive comparative experiments were conducted using a self-collected traffic signal light database. The results indicate that the proposed method, compared to traditional approaches, effectively handles interference factors such as lighting conditions, background variations, and changes in viewing angles. It achieves high accuracy and practicality in traffic signal light recognition, with an average precision rate of 93

Background:Heterogeneity is a critical characteristic of severe coronavirus disease 2019 (COVID-19) pneumonia. Integrating chest computed tomography (CT) imaging and plasma proteomics holds the potential to elucidate Image-Expression Axes (IEAs) that can effectively address this disease heterogeneity. Methods:A cohort of subjects diagnosed with severe COVID-19 pneumonia at 12 participating hospitals between December 2022 and March 2023 was prospectively screened for eligibility. Context-aware self-supervised representation learning (CSRL) was employed to extract intricate features from CT images. Quantification of plasma proteins was achieved using the Olink® inflammation panel. A deep learning model was meticulously trained, with CSRL features serving as input and the proteomic data as the target. This trained model facilitated the construction of IEAs, offering a representation of the underlying disease heterogeneity. The potential of these IEAs for prognostic and predictive enrichment was subsequently explored via conventional regression models. Results:The study cohort comprised 1979 eligible patients, who were stratified into a training set of 630 individuals and a testing set of 1349 individuals. Three distinct IEAs were identified: IEA1 was correlated with shock conditions, IEA2 was associated with the systemic inflammatory response syndrome (SIRS), and IEA3 was reflective of the coagulation profile. Notably, IEA1 (odds ratio [OR]= 0.52, 95 % confidence interval [CI]: 0.40 to 0.67, P < 0.001) and IEA2 (OR=0.74, 95 % CI: 0.62 to 0.90, P=0.002) exhibited significant associations with the risk of mortality. Intriguingly, patients characterized by lower IEA1 values (<-2, indicative of more severe shock) demonstrated a reduced mortality risk when administered with steroids. Conversely, patients with higher IEA2 values seemed to benefit from a judicious approach to fluid infusion. Conclusions:Our comprehensive approach, seamlessly integrating advanced deep learning techniques, proteomic profiling, and clinical data, has unraveled intricate interdependencies between IEAs, protein abundance patterns, therapeutic interventions, and ultimate patient outcomes in the context of severe COVID-19 pneumonia. These discoveries make a significant contribution to the rapidly advancing field of precision medicine, paving the way for tailored therapeutic strategies that can significantly impact patient care.

Macula fovea detection is a crucial molecular biological prerequisite for screening and diagnosing macular diseases. Without early detection and proper treatment, any abnormality involving the macula may lead to blindness. However, with the ophthalmologist shortage and time-consuming artificial evaluation, neither the accuracy nor effectiveness of the diagnosis process could be guaranteed. In this project, we proposed a light-weighted deep learning model based on ultra-widefield fundus (UWF) images for macula fovea detection tasks. This study collected 2300 ultra-widefield fundus images from Shenzhen Aier Eye Hospital in China. A light-weighted method based on a U-shape network (Unet) and Fully Convolution Network (FCN) approach is implemented on 1800 (before amplifying process) training fundus images, 400 (before amplifying process) validation images, and 100 test images. Three professional ophthalmologists were invited to mark the fovea. A method from the anatomy perspective is investigated. This approach is derived from the spatial relationship between the macula fovea and optic disc center in UWF. A set of parameters of this method is set based on the experience of ophthalmologists and verified to be effective. The ultra-widefield swept-source optical coherence tomography (UWF-OCT) approach is the grounded method. Through a comparison of proposed methods, we conclude that the proposed light-weighted Unet method outperformed other methods on macula fovea detection tasks.

近年来,随着以深度学习(DL)为代表的人工智能(AI)技术发展,为眼科领域带来了新的研究手段,提高了眼科疾病的筛查和诊断水平.目前,AI对糖尿病视网膜病变、白内障、早产儿视网膜病变、角膜炎等多种疾病的诊断效率较高.在青光眼方面,AI可用于分析眼底彩色照相、光学相干断层扫描(OCT)、视野等多模态影像综合评估结构及功能改变,从而提高青光眼的诊断水平.本文主要对AI在青光眼诊断中的研究进展进行综述,探讨其优势和现阶段的局限性.

Background:To comprehensively investigate the behaviors of oncologists with different working experiences and institute group styles in deep learning-based organs-at-risk (OAR) contouring. Methods:A deep learning-based contouring system (DLCS) was modeled from 188 CT datasets of patients with nasopharyngeal carcinoma (NPC) in institute A. Three institute oncology groups, A, B, and C, were included; each contained a beginner and an expert. For each of the 28 OARs, two trials were performed with manual contouring first and post-DLCS edition later, for ten test cases. Contouring performance and group consistency were quantified by volumetric and surface Dice coefficients. A volume-based and a surface-based oncologist satisfaction rate (VOSR and SOSR) were defined to evaluate the oncologists' acceptance of DLCS. Results:Based on DLCS, experience inconsistency was eliminated. Intra-institute consistency was eliminated for group C but still existed for group A and group B. Group C benefits most from DLCS with the highest number of improved OARs (8 for volumetric Dice and 10 for surface Dice), followed by group B. Beginners obtained more numbers of improved OARs than experts (7 v.s. 4 in volumetric Dice and 5 v.s. 4 in surface Dice). VOSR and SOSR varied for institute groups, but the rates of beginners were all significantly higher than those of experts for OARs with experience group significance. A remarkable positive linear relationship was found between VOSR and post-DLCS edition volumetric Dice with a coefficient of 0.78. Conclusions:The DLCS was effective for various institutes and the beginners benefited more than the experts.

As we know, recently deep learning networks have gained currency for some time under the background of the rise of big model, and it has been widely used for various areas including microbiology images recognition. Nowadays deep learning network models are also divided into many different types, and many new models are proposed every year to achieve better performance. After introducing their specific principles and composition structure, this paper compares three common different deep learning networks named Deep Neural Network notation (DNN), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) and introduces the recently emerging model named Residual network (Resnet), starting from a specific situation which calls for garbage classification. Moreover, during this experiment, it also gives the method V pipe a try which differs a lot from the former method and receives good results worth celebrating.

Compared with traditional fundus examination techniques, ultra-widefield fundus (UWF) images provide 200° panoramic images of the retina, which allows better detection of peripheral retinal lesions. The advent of UWF provides effective solutions only for detection but still lacks efficient diagnostic capabilities. This study proposed a retinal lesion detection model to automatically locate and identify six relatively typical and high-incidence peripheral retinal lesions from UWF images which will enable early screening and rapid diagnosis. A total of 24,602 augmented ultra-widefield fundus images with labels corresponding to 6 peripheral retinal lesions and normal manifestation labelled by 5 ophthalmologists were included in this study. An object detection model named You Only Look Once X (YOLOX) was modified and trained to locate and classify the six peripheral retinal lesions including rhegmatogenous retinal detachment (RRD), retinal breaks (RB), white without pressure (WWOP), cystic retinal tuft (CRT), lattice degeneration (LD), and paving-stone degeneration (PSD). We applied coordinate attention block and generalized intersection over union (GIOU) loss to YOLOX and evaluated it for accuracy, sensitivity, specificity, precision, F1 score, and average precision (AP). This model was able to show the exact location and saliency map of the retinal lesions detected by the model thus contributing to efficient screening and diagnosis. The model reached an average accuracy of 96.64%, sensitivity of 87.97%, specificity of 98.04%, precision of 87.01%, F1 score of 87.39%, and mAP of 86.03% on test dataset 1 including 248 UWF images and reached an average accuracy of 95.04%, sensitivity of 83.90%, specificity of 96.70%, precision of 78.73%, F1 score of 81.96%, and mAP of 80.59% on external test dataset 2 including 586 UWF images, showing this system performs well in distinguishing the six peripheral retinal lesions. Focusing on peripheral retinal lesions, this work proposed a deep learning model, which automatically recognized multiple peripheral retinal lesions from UWF images and localized exact positions of lesions. Therefore, it has certain potential for early screening and intelligent diagnosis of peripheral retinal lesions.

Fundus digital photography and optical coherence tomography (OCT) are currently the primary imaging approaches for early diagnosis and treatment of eye diseases. In recent years, the significant development in artificial intelligence (AI), particularly in machine learning (ML) and deep learning (DL) are new and vital technical-driven motivations impacting on the traditional diagnosis and treatment methods. At the same time, the ultra-wide field (UWF) imaging technology is getting widely accepted and prevalent by its obvious advantageous features of non-dilate pupils, express-track result and the vast pool of fundus viewing angles. As a result, numerous research have been done to explore AI in ultra-wide field fundus imaging ophthalmology for joint diagnosis and treatment. However, the current review of this method is still in least ink. We first outlines the application and impact of AI technology in ophthalmic diseases in the past ten years. With the following part exclusively summarizing the technical integration of ultra-wide field fundus images and AI technology in the past four years, which has brought innovations to clinical treatment methods for the diagnosis and treatment of ophthalmic diseases; finally, we analyzed the application and implementation of the novel technology as well as the potential limitations and challenges, to predict the possibility of the technology’s further principles role and values in clinical ophthalmology.

The subject addressed in this paper is identifying the recurrent faults from the recorded disturbance waveforms. The electrical equipment may experience self-clearing and instantaneous fault before the permanent faults occur, gradually reducing the insulation performance of the equipment. These faults are called incipient fault, which cannot be detected by the overcurrent relay for its low magnitude or short duration, but can occur many times before the permanent fault. At present, scholars have conducted related research of incipient faults. However, the causes of incipient fault may be different, it is difficult to apply these research methods to distribution fault early warning. Meanwhile, some incipient faults imply the high correlation and the same underlying causes. Such faults refer to recurrent faults . Therefore, recurrent faults identification should be finished before analyzing their cumulative effect on the permanent fault. Identification of these recurrent faults will facilitate distribution fault early warning and contributing to improving power supply reliability.

Macula fovea detection is a crucial prerequisite towards screening and diagnosing macular diseases. Without early detection and proper treatment, any abnormality involving the macula may lead to blindness. However, with the ophthalmologist shortage and time-consuming artificial evaluation, neither accuracy nor effectiveness of the diagnose process could be guaranteed. In this project, we proposed a deep learning approach on ultra-widefield fundus (UWF) images for macula fovea detection. This study collected 2300 ultra-widefield fundus images from Shenzhen Aier Eye Hospital in China. Methods based on U-shape network (Unet) and Fully Convolutional Networks (FCN) are implemented on 1800 (before amplifying process) training fundus images, 400 (before amplifying process) validation images and 100 test images. Three professional ophthalmologists were invited to mark the fovea. A method from the anatomy perspective is investigated. This approach is derived from the spatial relationship between macula fovea and optic disc center in UWF. A set of parameters of this method is set based on the experience of ophthalmologists and verified to be effective. Results are measured by calculating the Euclidean distance between proposed approaches and the accurate grounded standard, which is detected by Ultra-widefield swept-source optical coherence tomograph (UWF-OCT) approach. Through a comparation of proposed methods, we conclude that, deep learning approach of Unet outperformed other methods on macula fovea detection tasks, by which outcomes obtained are comparable to grounded standard method.

Aiming at the difference in the segmentation performance of the three segmentation target regions in the glioma image segmentation task based on the fully convolutional neural network, we propose a comprehensive evaluation method of neural network performance based on four evaluation indices. In addition, we analyze the performance and characteristics of neural network in the segmentation task of glioma, study the segmentation performance of neural network in the whole tumor (WT), tumor core (TC) and enhanced tumor (ET) regions, and propose a deep learning algorithm based on multiple networks in parallel. In this paper, the input image of the two-dimensional neural network is sliced, and the input of the three-dimensional neural network is processed in two ways: overlapping and non-overlapping, and in the image post-processing part, the three-dimensional image is reconstructed before the evaluation index is calculated. This article uses four evaluation indexes, which are Dice, Sensitivity, PPV, and Hausdorff, for the three segmentation target regions, and performs RSR* weight calculation, and finally performs a comprehensive evaluation. Experimental results show that Vnet has the best comprehensive segmentation performance, FCN-8s has the best segmentation performance in the TC area, Unet++ has the best segmentation performance in the ET area, and Vnet has the best segmentation performance in the WT area. Based on this, we propose a FUV multi-network parallel algorithm, combined with a reverse attention mechanism to improve the segmentation accuracy of the three segmentation target regions.