Aims Ischaemic heart disease (IHD) has been a significant public health issue worldwide. This study aims to predict the global burden of IHD in a timely and comprehensive manner. Methods and results Incidence, prevalence, deaths, and disability-adjusted life years (DALYs) for IHD from 1990 to 2021 were derived from the Global Burden of Disease 2021 database, and three models (linear, exponential, and Poisson regression) were used to estimate their trends over time at the global, regional, and national levels by age, sex, and country groups, with the gross domestic product per capita was applied to adjust the model. The model results revealed that the global burden of IHD is expected to increase continuously by 2050. By 2050, global IHD incidence, prevalence, deaths, and DALYs are projected to reach 67.3 million, 510 million, 16 million, and 302 million, respectively, which represents an increase of 116%, 106%, 80%, and 62%, respectively, from 2021. Moreover, the results showed that regions with lower sociodemographic index (SDI) bore a greater burden of IHD than those with higher SDI, with men having a higher burden of IHD than women. People over 70 years old account for a major part of the burden of IHD, and premature death of IHD is also becoming more serious. Conclusion The global burden of IHD will increase further by 2050, potentially due to population ageing and economic disparities. Hence, it is necessary to strengthen the prevention of IHD and formulate targeted strategies according to different SDI regions and special populations.

Early detection and treatment of congenital heart disease (CHD) can significantly improve the prognosis of children. However, inexperienced sonographers often face difficulties in recognizing CHD through transthoracic echocardiogram (TTE) images. In this study, 2-dimensional (2D) and Doppler TTEs of children collected from 2 clinical groups from Beijing Children's Hospital between 2018 and 2022 were analyzed, including views of apical 4 chamber, subxiphoid long-axis view of 2 atria, parasternal long-axis view of the left ventricle, parasternal short-axis view of aorta, and suprasternal long-axis view. A deep learning (DL) framework was developed to identify cardiac views, integrate information from various views and modalities, visualize the high-risk region, and predict the probability of the subject being normal or having an atrial septal defect (ASD) or a ventricular septaldefect (VSD). A total of 1,932 children (1,255 healthy controls, 292 ASDs, and 385 VSDs) were collected from 2 clinical groups. For view classification, the DL model reached a mean [SD] accuracy of 0.989 [0.001]. For CHD screening, the model using both 2D and Doppler TTEs with 5 views achieved a mean [SD] area under the receiver operating characteristic curve (AUC) of 0.996 [0.000] and an accuracy of 0.994 [0.002] for within-center evaluation while reaching a mean [SD] AUC of 0.990 [0.003] and an accuracy of 0.993 [0.001] for cross-center test set. For the classification of healthy, ASD, and VSD, the model reached the mean [SD] accuracy of 0.991 [0.002] and 0.986 [0.001] for within- and cross-center evaluation, respectively. The DL models aggregating TTEs with more modalities and scanning views attained superior performance to approximate that of experienced sonographers. The incorporation of multiple views and modalities of TTEs in the model enables accurate identification of children with CHD in a noninvasive manner, suggesting the potential to enhance CHD detection performance and simplify the screening process.

Background Stroke is a pressing public health issue worldwide. This study aimed to forecast the future temporal trends and distribution characteristics of the global stroke burden. Methods and Results Global data on stroke incidence, prevalence, deaths, and disability‐adjusted life years between 1990 and 2021 were obtained from the GBD (Global Burden of Disease) Study 2021. The temporal trends of stroke burden were estimated using various regression models. Gross domestic product per capita was adopted as a predictor in the model to consider the impact of economic development on stroke burden. Projections of stroke burden up to 2050 were generated using the optimal model selected based on the Akaike information criterion, encompassing global, World Bank income levels, national levels, and sex‐age groups. In 2050, we projected 21.43 million stroke cases, 159.31 million survivors, 12.05 million deaths, and 224.86 million disability‐adjusted life years due to stroke globally. From 2021 to 2050, there was a declining trend in the global age‐adjusted stroke rates, with −7% in incidence, −4% in prevalence, −28% in deaths, and −28% in disability‐adjusted life years. Upper‐middle‐income countries were projected to have the most severe stroke burden, followed by lower‐middle‐income countries, low‐income countries, and high‐income countries. The stroke burden in over half of the 204 countries and territories was expected to be alleviated from 2022 to 2050. Men and older women worldwide bear higher burden. Conclusions Stroke remains a serious global health challenge, especially in low‐income and middle‐income countries. Targeted implementation of prevention and interventions is imperative across diverse demographic groups.

Federated learning (FL) enables multiple client medical institutes collaboratively train a deep learning (DL) model with privacy protection. However, the performance of FL can be constrained by the limited availability of labeled data in small institutes and the heterogeneous (i.e., non-i.i.d.) data distribution across institutes. Though data augmentation has been a proven technique to boost the generalization capabilities of conventional centralized DL as a “free lunch”, its application in FL is largely underexplored. Notably, constrained by costly labeling, 3D medical segmentation generally relies on data augmentation. In this work, we aim to develop a vicinal feature-level data augmentation (VFDA) scheme to efficiently alleviate the local feature shift and facilitate collaborative training for privacy-aware FL segmentation. We take both the inner- and inter-institute divergence into consideration, without the need for cross-institute transfer of raw data or their mixup. Specifically, we exploit the batch-wise feature statistics (e.g., mean and standard deviation) in each institute to abstractly represent the discrepancy of data, and model each feature statistic probabilistically via a Gaussian prototype, with the mean corresponding to the original statistic and the variance quantifying the augmentation scope. From the vicinal risk minimization perspective, novel feature statistics can be drawn from the Gaussian distribution to fulfill augmentation. The variance is explicitly derived by the data bias in each individual institute and the underlying feature statistics characterized by all participating institutes. The added-on VFDA consistently yielded marked improvements over six advanced FL methods on both 3D brain tumor and cardiac segmentation.

BACKGROUND AND AIMS:Double-balloon endoscopy (DBE) is widely used in diagnosing small-bowel Crohn's disease (CD). However, CD misdiagnosis frequently occurs if inexperienced endoscopists cannot accurately detect the lesions. The CD evaluation may also be inaccurate owing to the subjectivity of endoscopists. This study aimed to use artificial intelligence (AI) to accurately detect and objectively assess small-bowel CD for more refined disease management. METHODS:We collected 28,155 small-bowel DBE images from 628 patients from January 2018 to December 2022. Four expert gastroenterologists labeled the images, and at least 2 endoscopists made the final decision with agreement. A state-of-the-art deep learning model, EfficientNet-b5, was trained to detect CD lesions and evaluate CD ulcers. The detection included lesions of ulcer, noninflammatory stenosis, and inflammatory stenosis. Ulcer grading included ulcerated surface, ulcer size, and ulcer depth. A comparison of AI model performance with endoscopists was performed. RESULTS:The EfficientNet-b5 achieved high accuracies of 96.3% (95% confidence interval [CI], 95.7%-96.7%), 95.7% (95% CI, 95.1%-96.2%), and 96.7% (95% CI, 96.2%-97.2%) for the detection of ulcers, noninflammatory stenosis, and inflammatory stenosis, respectively. In ulcer grading, the EfficientNet-b5 exhibited average accuracies of 87.3% (95% CI, 84.6%-89.6%) for grading the ulcerated surface, 87.8% (95% CI, 85.0%-90.2%) for grading the size of ulcers, and 85.2% (95% CI, 83.2%-87.0%) for ulcer depth assessment. CONCLUSIONS:The EfficientNet-b5 achieved high accuracy in detecting CD lesions and grading CD ulcers. The AI model can provide expert-level accuracy and objective evaluation of small-bowel CD to optimize the clinical treatment plans.

Background: In recent years, more and more people suffer from depression and anxiety. These symptoms are hard to be spotted and can be very dangerous. Currently, the Self-Reported Anxiety Scale (SAS) and Self-Reported Depression Scale (SDS) are commonly used for initial screening for depression and anxiety disorders. Howev-er, the information contained in these two scales is limited, while the symptoms of subjects are various and complex, which results in the inconsistency between the questionnaire evaluation results and the clinician's diagnosis results. To fully mine the scale data, we propose a method to extract the features from the facial expression and movements, which are generated from the video recorded simultaneously when subjects fill in the scale. Then we collect the facial expression, movements and scale information to establish a multimodal framework for improving the accuracy and robustness of the diagnosis of depression and anxiety.Methods: We collect the scale results of the subjects and the videos when filling in the scales. Given the two scales, SAS and SDS, we construct a model with two branches, where each branch processes the multimodal data of SAS and SDS, respectively. In the branch, we first build a convolutional neural network (CNN) to extracts the facial expression features in each frame of images. Secondly, we establish a long short-term memory (LSTM) network to further embedding the facial expression feature and build the connections between various frames, so that the movement feature in the video can be generated. Thirdly, we transform the scale scores into one-hot format, and feed them into the corresponding branch of the network to further mining the information of the multimodal data. Finally, we fuse the embeddings of these two branches to generate inference results of depression and anxiety.Results and conclusions: Based on the score results of SAS and SDS, our multimodal model further mines the video information, and can reach the accuracy of 0.946 in diagnosing depression and anxiety. This study demonstrates the feasibility of using our CNN-LSTM-based multimodal model for initial screening and diagnosis of depression and anxiety disorders with high diagnostic performance.

Automated echocardiogram interpretation with artificial intelligence (AI) has the potential to facilitate the serial diagnosis of heart defects by primary clinician. However, the fully automated and interpretable analysis pipeline for suggesting a treatment plan is largely underexplored. The present study targets to build an automatic and interpretable assistant for the transthoracic echocardiogram- (TTE-) based assessment of atrial septal defect (ASD) with deep learning (DL). We developed a novel deep keypoint stadiometry (DKS) model, which learns to precisely localize the keypoints, i.e., the endpoints of defects and followed by the absolute distance measurement with the scale. The closure plan and the size of the ASD occluder for transcatheter closure are derived based on the explicit clinical decision rules. A total of 3,474 2D and Doppler TTE from 579 patients were retrospectively collected from two clinical groups. The accuracy of closure classification using DKS (0.9425 ± 0.0052) outperforms the "black-box" model (0.7646 ± 0.0068; p < 0.0001) for within-center evaluation. The results in cross-center cases or using the quadratic weighted kappa as an evaluation metric are consistent. The fine-grained keypoint label provides more explicit supervision for network training. While DKS can be fully automated, clinicians can intervene and edit at different steps of the process as well. Our deep learning keypoint localization can provide an automatic and transparent way for assessing size-sensitive congenital heart defects, which has huge potential value for application in primary medical institutions in China. Also, more size-sensitive treatment planning tasks may be explored in the future.