Parkinson’s disease is a chronic neurodegenerative condition accompanied by a variety of motor and non-motor clinical symptoms. Diagnosing Parkinson’s disease presents many challenges, such as excessive reliance on subjective scale scores and a lack of objective indicators in the diagnostic process. Developing efficient and convenient methods to assist doctors in diagnosing Parkinson’s disease is necessary. In this paper, we study the skeleton sequences obtained from gait videos of Parkinsonian patients for early detection of the disease. We designed a Transformer network based on feature tensor fusion to capture the subtle manifestations of Parkinson’s disease. Initially, we fully utilized the distance information between joints, converting it into a multivariate time series classification task. We then built twin towers to discover dependencies within and across sequence channels. Finally, a tensor fusion layer was employed to integrate the features from both towers. In our experiments, our model demonstrated superior performance over the current state-of-the-art algorithm, achieving an 86.8% accuracy in distinguishing Parkinsonian patients from healthy individuals using the PD-Walk dataset.
Parkinson’s disease (PD) is a common neurodegenerative disorder, and accurate identification of PD is critical for clinical diagnosis and disease management. Electroencephalography (EEG) sensors provide reliable real-time brain signal acquisition, making them practical biosensing modalities for PD detection. However, due to their non-stationarity, single time-domain or frequency-domain analysis methods are insufficient to extract robust discriminative features from EEG signals. To address this challenge, we propose a multi-domain feature fusion EEG classification model, termed Multi-Domain Fusion Network (MDF-Net), which jointly integrates temporal, frequency-domain, and wavelet-domain representations for accurate PD recognition. MDF-Net employs a Temporal Attention-enhanced Temporal Convolutional Network (TTCN) to capture temporal dependencies and incorporates an improved 1D Convolutional Neural Network mixer module (Cmix) for multi-channel feature fusion. We constructed an EEG dataset of 415 subjects (289 healthy controls and 126 PD patients). Under 5-CV, the proposed method achieved a classification accuracy of 92.3%, an F1-score of 87.3%, and an AUC of 0.943. Experimental results demonstrate that multi-domain feature fusion effectively improves PD detection performance, and EEG sensor-based analysis shows strong potential for clinical application. This study provides a methodological reference for developing objective, practical computer-aided diagnostic tools for PD.
The annual increase in the incidence of Parkinson’s disease (PD) underscores the critical need for effective detection methods and devices. Gait video features based on camera sensors, as a crucial biomarker for PD, are well-suited for detection and show promise for the development of portable devices. Consequently, we developed a single-step segmentation method based on Savitzky–Golay (SG) filtering and a sliding window peak selection function, along with a Cross-Attention Fusion with Mamba-2 and Self-Attention Network (CMSA-Net). Additionally, we introduced a loss function based on Maximum Mean Discrepancy (MMD) to further enhance the fusion process. We evaluated our method on a dual-view gait video dataset that we collected in collaboration with a hospital, comprising 304 healthy control (HC) samples and 84 PD samples, achieving an accuracy of 89.10% and an F1-score of 81.11%, thereby attaining the best detection performance compared with other methods. Based on these methodologies, we designed a simple and user-friendly portable PD detection device. The device is equipped with various operating modes—including single-view, dual-view, and prior information correction—which enable it to adapt to diverse environments, such as residential and elder care settings, thereby demonstrating strong practical applicability.
We propose Heterogeneous Dual-path Contrastive Architecture (HDCA) for action recognition. Our model involves a spatial pathway and a temporal pathway; these two pathways employ distinct backbone networks and input formats, tailored to the specific properties of spatial features and temporal features. The spatial pathway processes super images to capture spatial semantics while the temporal pathway operates on frame sequences to model motion patterns. This targeted design can precisely capture the scenes and motions depicted in videos while improving parameter efficiency. To establish a cross-modality complementary enhancement mechanism, we develop cross-modality contrastive loss and intra-group contrastive loss to train the HDCA. These contrastive losses work synergistically to maximize the similarity of feature representations among videos belonging to the same class while minimizing similarity across different classes, achieving cross-modality alignment through cross-modality contrastive loss and enhancing intra-group compactness via intra-group contrastive loss. HDCA fully exploits the complementary strengths of spatial features and temporal features in action recognition. Systematic experiments on three benchmark datasets validate the effectiveness and superiority of our approach, which support the motivation and hypothesis of our model design. The experimental results demonstrate that our model achieves competitive performance compared to existing state-of-the-art approaches for action recognition. Notably, performance gains increase with dataset complexity, indicating that discriminative correlation information between modalities learned by HDCA yield greater performance gains in the recognition tasks of complex videos.
Parkinson’s disease (PD) is associated with characteristic gait impairment, motivating objective screening methods based on biomechanical signals. This study presents a lightweight, physics-informed framework for PD gait screening using ground reaction force (GRF) signals acquired from force plates, together with a prototype acquisition-and-analysis system for practical screening workflows. Continuous GRF recordings are segmented into complete gait cycles, from which bilateral physics-informed features are constructed, including normalized force, dynamics-derived acceleration and velocity, and friction-related descriptors. The resulting feature tensor is then standardized and used as input to a Feature-Grouped Gated Fusion Network (FGGF-Net). The proposed model separately encodes force–acceleration features and velocity–ratio features using low-order nonlinear and linear pathways, respectively, and integrates them via gated fusion with a residual baseline pathway. Under subject-wise five-fold cross-validation, FGGF-Net achieves strong subject-level performance, reaching 94.8% accuracy, 92.9% F1-score, and 0.974 AUC, while consistently outperforming representative baselines. Ablation studies further verify the effectiveness of feature grouping and gated fusion. In addition, the trained model remains compact (1.09 M parameters, 4.16 MB) and supports millisecond-level CPU inference, making the proposed framework promising for practical force-plate screening workflows.