Structural magnetic resonance imaging (sMRI) is commonly used for the identification of Alzheimer's disease because of its keen insight into atrophy-induced changes in brain structure. Current mainstream convolutional neural network-based deep learning methods ignore the long-term dependencies between voxels; thus, it is challenging to learn the global features of sMRI data. In this study, an advanced deep learning architecture called Brain Informer (BraInf) was developed based on an efficient self-attention mechanism. The proposed model integrates representation learning, feature distilling, and classifier modeling into a unified framework. First, the proposed model uses a multihead ProbSparse self-attention block for representation learning. This self-attention mechanism selects the first ⌊lnN⌋ elements that can represent the overall features from the perspective of probability sparsity, which significantly reduces computational cost. Subsequently, a structural distilling block is proposed that applies the concept of patch merging to the distilling operation. The block reduces the size of the three-dimensional tensor and further lowers the memory cost while preserving the original data as much as possible. Thus, there was a significant improvement in the space complexity. Finally, the feature vector was projected into the classification target space for disease prediction. The effectiveness of the proposed model was validated using the Alzheimer's Disease Neuroimaging Initiative dataset. The model achieved 97.97% and 91.89% accuracy on Alzheimer's disease and mild cognitive impairment classification tasks, respectively. The experimental results also demonstrate that the proposed framework outperforms several state-of-the-art methods.

Pansharpening is a very debated spatio-spectral fusion problem. It refers to the fusion of a high spatial resolution panchromatic (PAN) image with a lower spatial but higher spectral resolution multispectral (LRMS) image in order to obtain an image with high resolution in both the domains. In this paper, we propose a novel variational optimization-based (VO) approach to address this issue incorporating the outcome of a deep convolutional neural network (DCNN). This solution can take advantages of both the paradigms. On one hand, higher performance can be expected introducing machine learning methods based on the training by examples philosophy into VO approaches. On other hand, the combination of VO techniques with DCNNs can aid the generalization ability of these latter. In particular, we formulate a l2-based proximal deep injection term to evaluate the distance between the DCNN outcome and the desired high spatial resolution multispectral image. This represents the regularization term for our VO model. Furthermore, a new data fitting term measuring the spatial fidelity is proposed. Finally, the proposed convex VO problem is efficiently solved by exploiting the framework of the alternating direction method of multipliers, thus guaranteeing the convergence of the algorithm. Extensive experiments both on simulated and real datasets demonstrate that the proposed approach can outperform state-of-the-art spatio-spectral fusion methods, even showing a significant generalization ability.

Background: Nowadays, intelligent medicine is gaining widespread attention, and great progress has been made in Western medicine with the help of artificial intelligence to assist in decision making. Compared with Western medicine, traditional Chinese medicine (TCM) involves selecting the specific treatment method, prescription, and medication based on the dialectical results of each patient’s symptoms. For this reason, the development of a TCM-assisted decision-making system has lagged. Treatment based on syndrome differentiation is the core of TCM treatment; TCM doctors can dialectically classify diseases according to patients’ symptoms and optimize treatment in time. Therefore, the essence of a TCM-assisted decision-making system is a TCM intelligent, dialectical algorithm. Symptoms stored in electronic medical records are mostly associated with patients’ diseases; however, symptoms of TCM are mostly subjectively identified. In general electronic medical records, there are many missing values. TCM medical records, in which symptoms tend to cause high-dimensional sparse data, reduce algorithm accuracy. Objective: This study aims to construct an algorithm model compatible for the multidimensional, highly sparse, and multiclassification task of TCM syndrome differentiation, so that it can be effectively applied to the intelligent dialectic of different diseases. Methods: The relevant terms in electronic medical records were standardized with respect to symptoms and evidence-based criteria of TCM. We structuralized case data based on the classification of different symptoms and physical signs according to the 4 diagnostic examinations in TCM diagnosis. A novel cross-feature generation by convolution neural network model performed evidence-based recommendations based on the input embedded, structured medical record data. Results: The data set included 5273 real dysmenorrhea cases from the Sichuan TCM big data management platform and the Chinese literature database, which were embedded into 60 fields after being structured and standardized. The training set and test set were randomly constructed in a ratio of 3:1. For the classification of different syndrome types, compared with 6 traditional, intelligent dialectical models and 3 click-through-rate models, the new model showed a good generalization ability and good classification effect. The comprehensive accuracy rate reached 96.21%. Conclusions: The main contribution of this study is the construction of a new intelligent dialectical model combining the characteristics of TCM by treating intelligent dialectics as a high-dimensional sparse vector classification task. Owing to the standardization of the input symptoms, all the common symptoms of TCM are covered, and the model can differentiate the symptoms with a variety of missing values. Therefore, with the continuous improvement of disease data sets, this model has the potential to be applied to the dialectical classification of different diseases in TCM.

Available evidence suggests that dynamic functional connectivity can capture time-varying abnormalities in brain activity in resting-state cerebral functional magnetic resonance imaging (rs-fMRI) data and has a natural advantage in uncovering mechanisms of abnormal brain activity in schizophrenia (SZ) patients. Hence, an advanced dynamic brain network analysis model called the temporal brain category graph convolutional network (Temporal-BCGCN) was employed. Firstly, a unique dynamic brain network analysis module, DSF-BrainNet, was designed to construct dynamic synchronization features. Subsequently, a revolutionary graph convolution method, TemporalConv, was proposed based on the synchronous temporal properties of features. Finally, the first modular test tool for abnormal hemispherical lateralization in deep learning based on rs-fMRI data, named CategoryPool, was proposed. This study was validated on COBRE and UCLA datasets and achieved 83.62% and 89.71% average accuracies, respectively, outperforming the baseline model and other state-of-the-art methods. The ablation results also demonstrate the advantages of TemporalConv over the traditional edge feature graph convolution approach and the improvement of CategoryPool over the classical graph pooling approach. Interestingly, this study showed that the lower-order perceptual system and higher-order network regions in the left hemisphere are more severely dysfunctional than in the right hemisphere in SZ, reaffirmings the importance of the left medial superior frontal gyrus in SZ. Our code was available at: https://github.com/swfen/Temporal-BCGCN.

In recent years, sparse unmixing has attracted significant attention as it can effectively avoid the bottleneck problems associated with the absence of pure pixels and the estimation of the number of endmembers in hyperspectral scenes. The joint-sparsity model has outperformed the single sparse unmixing method. However, the joint-sparsity model might cause some aliasing artifacts for the pixels on the boundaries of different constituent endmembers. To address this shortcoming, researchers have developed many unmixing algorithms based on low-rank representation, which makes good use of the global structure of data. In addition, the high mutual coherence of spectral libraries strongly affects the applicability of spare unmixing. In this study, adopting combined constraints imposing sparsity and low rankness, a novel algorithm called non-convex joint-sparsity and low-rank unmixing with dictionary pruning (NCJSpLRUDP) is developed. In particular, we impose sparsity on the abundance matrix using the l_2,p mixed norm, and we also employ the weighted Schatten p-norm instead of the convex nuclear norm as an approximation for the rank. The key parameter p is set between 0.4 and 0.6, and a good quality sparse solution is generated. The effectiveness of the proposed algorithm is demonstrated on both simulated and real hyperspectral datasets.