The hierarchical factor structure of the PID-5-BF+M inventory proved consistent in its application to older adults. Consistent internal structure was noted for both the domain and facet scales. The CD-RISC assessment yielded logically related findings. The domain of Negative Affectivity, including Emotional Lability, Anxiety, and Irresponsibility, exhibited a negative relationship with the concept of resilience.
According to the outcomes of this study, the construct validity of the PID-5-BF+M in senior citizens is substantiated. Further investigation into the instrument's age-neutral qualities is still required, however.
The results of this study unequivocally support the construct validity of the PID-5-BF+M inventory in older adults. Further research into the instrument's effectiveness irrespective of age is still required.
To guarantee secure operation of power systems, simulation analysis is crucial for pinpointing possible hazards. Large-disturbance rotor angle stability and voltage stability frequently present interconnected challenges within operational contexts. Accurately identifying the predominant instability mode (DIM) between them is vital for the formulation of targeted power system emergency control actions. However, the process of identifying DIMs has invariably relied upon the expertise and experience of human specialists. Utilizing active deep learning (ADL), this article proposes a novel DIM identification framework to distinguish among stable operation, rotor angle instability, and voltage instability. To streamline the labeling process for the DIM dataset when constructing deep learning models, a two-stage batch-mode integrated active learning approach, encompassing pre-selection and clustering, is designed for the platform. In each iteration, it chooses only the most valuable samples for labeling, focusing on both the information they contain and their diversity to enhance query effectiveness, resulting in a considerable reduction in the amount of labeled samples required. Comparative analysis of the proposed approach against conventional methods on the CEPRI 36-bus system and the Northeast China Power System reveals superior accuracy, label efficiency, scalability, and adaptability to fluctuating operational conditions.
Feature selection tasks are facilitated by the embedded feature selection approach, which leverages a pseudolabel matrix to guide the subsequent learning of the projection matrix (selection matrix). The pseudo-label matrix learned through spectral analysis from a relaxed problem interpretation has a certain degree of divergence from actual reality. In order to resolve this issue, we formulated a feature selection framework, drawing principles from classical least-squares regression (LSR) and discriminative K-means (DisK-means), and named it the fast sparse discriminative K-means (FSDK) feature selection method. To preclude a trivial solution arising from unsupervised LSR, a weighted pseudolabel matrix incorporating discrete traits is introduced initially. selleck This condition allows for the removal of any constraints placed on the pseudolabel matrix and the selection matrix, considerably simplifying the combinatorial optimization problem. Secondly, a l2,p-norm regularizer is implemented to ensure the row sparsity of the selection matrix, offering adaptable p-values. The FSDK model, a novel feature selection framework, is thus constructed by integrating the DisK-means algorithm and l2,p-norm regularization, with the aim of optimizing sparse regression problems. Furthermore, our model exhibits a linear correlation with the number of samples, facilitating the swift handling of extensive datasets. In-depth tests on various data sets clearly demonstrate the strength and expediency of FSDK.
Kernelized maximum-likelihood (ML) expectation maximization (EM) methods, spurred by the kernelized expectation maximization (KEM) approach, have emerged as a powerful tool in PET image reconstruction, demonstrating superior performance to numerous previous state-of-the-art techniques. The methods, though advantageous, suffer from the common problems of non-kernelized MLEM methods: potentially high reconstruction variance, sensitivity to the number of iterations, and the inherent challenge of simultaneously maintaining image details and suppressing image fluctuations. Using data manifold and graph regularization approaches, this paper designs a novel regularized KEM (RKEM) method for PET image reconstruction, with a kernel space composite regularizer. In the composite regularizer, a convex kernel space graph regularizer smooths kernel coefficients, a concave kernel space energy regularizer amplifies their energy, and a composition constant is analytically fixed to guarantee the convexity of the final regularizer. Employing a composite regularizer, PET-only image priors are readily implemented to address the difficulty of KEM, stemming from the incompatibility between MR priors and the PET images. Through the use of a kernel space composite regularizer and optimization transfer, a globally convergent iterative RKEM reconstruction algorithm is derived. The proposed algorithm's superiority over KEM and conventional methods is established through a comprehensive evaluation of simulated and in vivo data, encompassing tests and comparisons.
List-mode PET image reconstruction is indispensable for PET scanners equipped with numerous lines-of-response and enhanced by the inclusion of information regarding time-of-flight and depth-of-interaction. Despite the potential of deep learning, its implementation in list-mode PET image reconstruction has not advanced, primarily because list data is composed of a sequence of bit codes, making it incompatible with the processing power of convolutional neural networks (CNNs). This research presents a novel list-mode PET image reconstruction method, incorporating the deep image prior (DIP), an unsupervised convolutional neural network. This initial integration of list-mode PET and CNNs for image reconstruction is detailed here. Employing an alternating direction method of multipliers, the LM-DIPRecon method, which is a list-mode DIP reconstruction technique, alternately applies the regularized list-mode dynamic row action maximum likelihood algorithm (LM-DRAMA) and the MR-DIP. We compared LM-DIPRecon against LM-DRAMA, MR-DIP, and sinogram-based DIPRecon methods, using both simulated and real clinical data, and found LM-DIPRecon to produce sharper images and better contrast-noise tradeoff curves. urinary metabolite biomarkers The LM-DIPRecon's performance in quantitative PET imaging with limited events highlights its usefulness and the accuracy of preserved raw data. Consequently, the greater temporal detail in list data in comparison to dynamic sinograms makes list-mode deep image prior reconstruction a promising approach for improving both 4D PET imaging and motion correction procedures.
For the past few years, research heavily leveraged deep learning (DL) techniques for the analysis of 12-lead electrocardiogram (ECG) data. medical demography Yet, the assertion of deep learning's (DL) superiority to traditional feature engineering (FE) approaches, rooted in domain understanding, remains uncertain. Consequently, whether the fusion of deep learning with feature engineering may outperform a single-modality method remains ambiguous.
In light of the existing research voids and recent substantial experiments, we re-examined three tasks: cardiac arrhythmia diagnosis (multiclass-multilabel classification), atrial fibrillation risk prediction (binary classification), and age estimation (regression). Our training encompassed 23 million 12-lead ECG recordings, which served as the foundational dataset for three distinct models per task: i) a random forest, inputting feature extraction (FE); ii) a comprehensive deep learning (DL) model; and iii) a fusion model encompassing both feature extraction (FE) and deep learning (DL).
In the classification tasks, FE demonstrated results equivalent to DL, but with substantially reduced data requirements. For the regression task, DL's performance was superior to that of FE. Integration of the front end with deep learning did not provide enhanced performance compared to using deep learning alone. These findings received corroboration from the supplementary PTB-XL dataset.
Our study found that deep learning (DL) offered no substantial improvement over feature engineering (FE) for traditional 12-lead ECG-based diagnosis, whereas it significantly enhanced performance on non-traditional regression tasks. Our findings revealed no improvement when incorporating FE into DL compared to DL alone. This indicates that the features learned by FE were redundant with those learned by the deep learning model.
Our study's outcome highlights key recommendations for machine-learning strategies and data guidelines for the purpose of 12-lead ECG processing. To achieve peak performance, in cases of non-standard tasks with readily available large datasets, deep learning is the preferred method. For tasks characterized by classical methodologies and involving minimal data, a feature engineering technique might represent the superior selection.
Significant implications arise from our findings, focusing on optimal machine learning strategies and data handling practices for 12-lead ECG analysis in diverse contexts. If the pursuit of optimal performance involves a nontraditional task with a vast dataset, deep learning proves to be the optimal method. For tasks of a traditional nature and/or datasets of a small size, a feature engineering method might be a preferable choice.
Employing both mix-up and adversarial training, this paper presents a novel method, MAT-DGA, for domain generalization and adaptation in myoelectric pattern recognition, aimed at resolving the challenge of cross-user variability.
A unified framework encompassing both domain generalization (DG) and unsupervised domain adaptation (UDA) is constructed using this method. The DG procedure extracts user-neutral data from the source domain to build a model suitable for a new user in a target domain. The UDA method then further improves the model's proficiency with a few unlabeled examples supplied by this new user.