Chest X-Ray Vision-Language pretraining (VLP) leverages large-scale radiograph-report pairs to develop joint image-text representations, demonstrating significant potential for medical image diagnosis. However, existing VLP approaches often overlook the multi-view nature of chest X-Rays, and some multi-view methods apply uniform feature fusion, neglecting view-key semantic contributions. Moreover, random cross-modal Masked Language Modeling (MLM) fails to facilitate effective interactions, impeding representation alignment. Additionally, global alignment in VLP may lead to the false-negative problem. To address these limitations, we propose a novel medical VLP framework comprising three core components. First, a Key Semantics-enhanced Multi-view MLM module aggregates pathology-relevant patches across views, providing semantically rich supervision for MLM. A local semantics enhancing approach, which identifies and aggregates pathology-relevant key patches across views to guide MLM. Second, a Frontal-Lateral Alignment module extracts view-specific pathological features, ensuring semantic consistency and preserving critical information during aggregation. This module independently extracts pathological features from both views to preserve view-specific information while ensuring semantic consistency, which mitigates the loss of crucial information during aggregation. Third, a High-order Semantic Alignment approach mitigates false-negative issues by aligning features with semantically consistent clusters, enhancing global alignment through prototype-level semantics. Extensive experiments across seven public datasets demonstrate that our framework outperforms state-of-the-art methods in four downstream tasks, validating its efficacy. The code is available at https://github.com/sajiutea/F-L.

Brain age has emerged as a critical biomarker for assessing neurodevelopmental health and aging trajectories, demonstrating significant potential in early detection and monitoring of neurological and psychiatric disorders. While existing brain age prediction models predominantly rely on structural MRI (sMRI) due to their rich anatomical detail and predictive accuracy, resting-state functional MRI (rs-fMRI)—which captures dynamic brain connectivity—remains underutilized despite offering complementary insights. This study proposes a highly accurate and generalizable brain age prediction framework that effectively fuses sMRI and rs-fMRI modalities. Specifically, we integrate a Squeeze-and-Excitation Transformer for structural feature extraction with a Graph Frequency Recurrent Network for modeling functional dynamics. Our hybrid model achieves a MAE of 1.31 years and Pearson's R of 0.976 on the ABIDE I dataset while generalizing effectively across independent datasets, thus demonstrating the utility of multimodal fusion for robust brain age estimation.

Bot accounts on microblogging platforms significantly impact information reliability and cyberspace security. Accurately identifying these bots is essential for effective community governance and opinion management. This article introduces a category of online social behavior features (OSBF), derived from microblog behaviors such as emotional expression, language organization, and self-description. Through a series of experiments, OSBF has demonstrated the stable and robust performance in characterizing and detecting microblog bots on Twitter and Chinese Weibo. By identifying significant differences in OSBF between bot and human accounts, we established an OSBF-based detection model. This model showed excellent performance across multitask and multiscale challenges in two English Twitter datasets. Additionally, we explored cross-language and cross-dataset applications using two Chinese Weibo datasets, further affirming the model's effectiveness and robustness. The experimental results confirm that our OSBF-based model surpasses existing methods in detecting microblog bots.

Diffusion models show great potential in solving inverse problems, including MRI reconstruction. With its unique characteristics, medical imaging demands both efficiency and accuracy in the reconstruction process. However, existing MRI reconstruction methods based on diffusion models often fall short of fully leveraging the available measurements during sampling. Consequently, these methods suffer from compromised reconstruction quality and elevated bias, especially when dealing with large acceleration factors. In response to these challenges, we propose Dual Manifold Constraints (DMC), a fast MRI reconstruction method based on diffusion models. We treat the sampling process as a combination of denoising and adding noise processes, and we constrain these two processes using both pristine measurements and their noisy counterparts to adapt to the geometry of diffusion. It’s worth noting that we propose a method to estimate the noisy measurement that satisfies the sub-sampling process to maintain the current data manifold when performing data consistency constraints. Experimental results show that our method outperforms the latest diffusion-based methods regarding both reconstruction speed and accuracy, and exhibits strong out-of-distribution generalization performance.

With the explosive popularity of social media, more and more people, including those with depressive symptoms, are starting to express their emotions online through vlogs recently, which makes it important for video-based depression recognition. As video data contains rich acoustical and visual information, the main challenges faced by existing methods include (1) how to accurately mine features associated with depression in massive data and (2) how to effectively fuse various features from different modalities. In this paper, a multi-domain acoustical-visual information fusion network (MDAVIF) is designed to extract depressive spatio-temporal features from image sequences and audios, and an adaptive feature interaction module is proposed to mix these features. Combined with two autoencoders to retain information and prevent overfitting, the proposed method obtains the state-of-the-art result with the precision of 74.25% and the F1-Score of 75.25% when evaluated on the D-vlog dataset.

With the rise of large-scale model applications, leveraging these models as the base network for efficient transfer learning has garnered increasing attention. Currently, parameter-efficient transfer learning methods have made significant improvements in reducing the number of trainable parameters but introduce latency during inference. In this study, we propose an enhanced adaptation of the adapter using a reparameterization technique, revamping the activating layers into linear layers. This modification retains the high-dimensional fine-tuning capability of the adapter for visual tasks while avoiding additional inference latency. We name this plug-and-play module the Re3adapter, which optimizes the model with only 0.26% of the parameters and introduces no inference latency. Experimental results demonstrate its clear advantages in traditional classification and medical tasks.

The brain age is one of important biomarkers for identifying neurodegenerative diseases. Most existing works construct prediction models based on brain MRIs by deep learning methods while multi-scale features of MRI have been paid less attention. So, this paper proposes a new brain age predicting model, ERU-Net, which introduces an adaptive feature fusion (AFF) block into U-Net, for exploiting multi-scale features for high estimating accuracy. Moreover, the attention gate is also applied to enhance this model’s ability to extract important features among various brain regions. On a dataset of 6318 healthy people, ERU-Net achieves MAE = 2.85 and Pearson correlation coefficient R = 0.98, which outperforms other existing models for brain age estimation. In addition, by the results on the neurodegenerative disease datasets, it is clear that for the index of brain age gap (BAG), there are significantly statistical differences between healthy subjects and neurodegenerative patients such as Aizheimer’s disease, mild cognitive impairment, schizophrenia and Parkinson disease.

With the increasing number of depressed patients and the development of computer vision technology, the study of individual automatic depression estimation (ADE) methods based on facial images has attracted much attention in recent years. Most existing works focus on obtaining informative features from the whole images with advancing deep learning models, while the local tiny changes and the fusion of different features have been paid less attention. In this paper, a two-branch predicting model with a elaborate transfomer block NFFT is designed to combine global and local features extracted from whole images and image patches for predicting the depression score precisely. Besides, a classification head is added to guide the regression results for improving accuracy. Experiment results on the AVEC2014 dataset (MAE=5.81, RMSE=7.49) demonstrate that the proposed model outperforms other methods, and the extended experiments on one new dataset (CCPL) are conducted to validate the generalizability and robustness of our model.

The aging process of human brain is complex, which can result in brain structural changes. One promising way to gain a deep understanding of aging process is using machine learning, typically convolutional neural network (CNN), to predict brain age based on magnetic resonance imaging data. Though CNN has a strong ability to capture features from a small local region of the input image, it lacks the ability to capture global features of the surrounding neighbors. Thus, in this paper, we propose the squeeze and excitation transformer (SQET) for pursuing high-accuracy brain age estimation, in which a squeeze and excitation module is designed and fused in conventional self-attention in the transformer structure to capture global features among different localities even if they are spatially far apart. In particular, for 9 public datasets with 6,318 healthy brain Tl-MRIs with an age range of 6-88, our proposed SQET can achieve the result of 2.55 MAE and the correlation coefficient r=0.983, which has significantly outperformed all other reported models up to now.

The brain age, estimated from the brain MRI data, is found to be a promising biomarker for human brain development and neuroanatomical aging processes. A well-performed brain age predicting model is in great demand for many applications like healthcare and disease diagnosis. In this paper, we proposed a dimensional-attention-based 3D convolutional neural network (DACNN) to estimate the biological age for developing normal brain from T1-weighted MRI, in which a dimensional attention module was designed and applied to restrain noises and increase the weights of effective voxels for feature maps. Experimental results indicated that our model significantly outperformed the best reported methods up to now. In particular, with the dilated convolution, the proposed DACNN achieved the state-of-the-art result of 1.01 MAE on a combined dataset consisted of 880 healthy children and adolescents.

Early detection of depression is important to improve human well-being. This paper proposes a new method to detect depression through time-frequency analysis of Internet behaviors. We recruited 728 postgraduate students and obtained their scores on a depression questionnaire (Zung Self-rating Depression Scale, SDS) and digital records of Internet behaviors. By time-frequency analysis, classification models are built to differentiate higher SDS group from lower group, and prediction models are built to identify mental status of depressed group more precisely. Experimental results show classification and prediction models work well, and time-frequency features are effective in capturing the changes of mental health status. Results of this paper are useful to improve the performance of public mental health services.

Automatic emotion recognition from gaits information is discussed in this paper, which has been investigated widely in the fields of human-machine interaction, psychology, psychiatry, behavioral science, etc. The gaits information is non-contact, collected from Microsoft kinects, and contains 3-dimensional coordinates of 25 joints per person. These joints coordinates vary with the time. So, by the discrete Fourier transform and statistic methods, some time-frequency features related to neutral, happy and angry emotion are extracted and used to establish the classification model to identify these three emotions. Experimental results show this model works very well, and time-frequency features are effective in characterizing and recognizing emotions for this non-contact gait data. In particular, by the optimization algorithm, the recognition accuracy can be further averagely improved by about 13.7 percent.

Suicide is becoming a serious problem, and how to prevent suicide has become a very important research topic. The development of Social Network System (SNS) provides an ideal platform to monitor persons' suicidal ideation. Based on Sina microblog (Weibo), this paper proposes a real-time monitoring system detecting users' suicidal ideation. From 59046 posts collected with labels of either suicide or non-suicide, we extract new features based on content and emotion. Finally, four different classifiers including Support Vector Machine(SVM), Multinomial Naive Bayes(MultiNB), Logistic Regression(LR), Multi Layer Perception(MLP) are used to construct classified model respectively. Experimental results show that it is possible to detect suicidal ideation from microblogs, and the result of MLP is the best, whose F1 is up to 67.6%.

Early detection of depression is important to improve human well-being. This paper proposes a new method to detect depression through time-frequency analysis of Internet behaviors. We recruited 728 postgraduate students and obtained their scores on a depression questionnaire (Zung Self-rating Depression Scale, SDS) and digital records of Internet behaviors. By time-frequency analysis, we built classification models for differentiating higher SDS group from lower group and prediction models for identifying mental status of depressed group more precisely. Experimental results show classification and prediction models work well, and time-frequency features are effective in capturing the changes of mental health status. Results of this paper might be useful to improve the performance of public mental health services.

By incorporating diverse sources of evidence of relevance, learning to rank has been widely applied to real-time Twitter search, where users are interested in fresh relevant messages. Such approaches usually rely on a set of training queries to learn a general ranking model, which we believe that the benefits brought by learning to rank may not have been fully exploited as the characteristics and aspects unique to the given target queries are ignored. In this paper, we propose to further improve the retrieval performance of learning to rank for real-time Twitter search, by taking the difference between queries into consideration. In particular, we learn a query-biased ranking model with a semi-supervised transductive learning algorithm so that the query-specific features, e.g. the unique expansion terms, are utilized to capture the characteristics of the target query. This query-biased ranking model is combined with the general ranking model to produce the final ranked list of tweets in response to the given target query. Extensive experiments on the standard TREC Tweets11 collection show that our proposed query-biased learning to rank approach outperforms strong baseline, namely the conventional application of the state-of-the-art learning to rank algorithms.

The parametrization for one kind of multifilter banks generating balanced multiwavelets is presented in this paper, in which two lowpass filters are flipping filters, and two highpass filters have linear phase. Based on these parametric expressions, some balanced multiwavelets and analysis-ready multiwavelets are constructed, which are symmetric, or antisymmetric. Moreover, on the basis of balanced multiwavelet transform, a new method of multiple description coding is given, and experiments show that this method works well. Compared with the traditional multiple description coding method, this method has low redundancy.

Balanced multiwavelets with interpolatory property are discussed in this paper. This kind of multiwavelets can have a sampling property like Shannon's sampling theorem. It has been shown that the corresponding matrix-valued refinable mask has special structure, and an orthogonal multifilter bank {H(z),G(z)} can be reduced to a scalar valued conjugate quadrature filter (CQF) a(z) . But it does not mean that any scalar CQF can form a "good" multifilter bank which can generate a vector-valued refinable function with some degree of smoothness. In the context of balanced multiwavelets, we give the definition of transferring balance order, which a scalar CQF a(z) satisfies, to guarantee that the multiwavelet Ψ generated is balanced. On the basis of the parametrization of a scalar CQF with any length and conditions of transferring balance order, parametrization of multifilter banks which can generate interpolatory multiwavelet and interpolatory scaling function, is gotten. Moreover, some balanced interpolatory multiwavelets have been constructed. Interpolatory analysis-ready multiwavelets (armlets) are also discussed in this paper. It is known that conditions of armlets are easy to validate, compared with balanced multiwavelets. But it will be present that if the corresponding scaling function Φ is interpolatory, the multiwavelet Ψ is balanced of order n if and only if it is an armlet of order n. Finally, the application of balanced multiwavelets with interpolatory property in image processing is also discussed.