Subcortical vascular cognitive impairment (SVCI) is a heterogeneous cognitive impairment caused by small vessel disease. Diagnosis of SVCI remains challenging when neuropsychological assessment is impractical. This study proposes a diffusion tensor imaging (DTI)-based DenseNet to identify SVCI from subcortical ischemic vascular disease (SIVD) and to profile multidomain cognitive risks. We collected neuropsychological scales and DTI from 134 SVCI and 171 SIVD patients in our internal dataset for model development. An external target-domain dataset of 90 SVCI and 103 SIVD patients was used for unsupervised domain adaptation (UDA). Within this dataset, 45 SVCI and 53 SIVD patients were used for unlabeled UDA fitting; the remaining 45 SVCI and 50 SIVD patients were held out as a target-domain test set. Model-generated salient maps identified white matter (WM) regions associated with SVCI. Mutual information (MI) maps between DTI and 6 neuropsychological scales were computed to identify structural correlates of cognitive domains for cognitive profiling. We computed structural similarity index measure (SSIM) between individual-level salient maps derived from DenseNet and the MI maps for unsupervised clustering to stratify domain-specific cognitive impairment risk in SVCI. The DenseNet achieves high accuracy (0.902 internal, 0.926 target-domain) with AUCs of 0.951 and 0.942, respectively. SVCI probabilities reflect cognitive severity, and salient maps are associated with neuropsychological performance. Regarding cognitive profiling, each cognitive domain is divided into low, moderate, and high subgroups, with significantly different SSIM. Our DTI-based study demonstrates accurate SVCI identification and individualized multi-domain cognitive profiling. This offers a complementary framework to support diagnosis and personalized intervention.

The integration of multiple electrophysiological biomarkers is crucial for monitoring neonatal seizure dynamics. The present study aimed to characterize the temporal dynamics of neonatal seizures by analyzing intrinsic waveforms of epileptic electroencephalogram (EEG) signals. We proposed a complementary set of methods considering envelope power, focal sharpness changes, and nonlinear patterns of EEG signals of 79 neonates with seizures. Features derived from EEG signals were used as input to the machine learning classifier. All three characteristics were significantly elevated during seizure events, as agreed upon by all viewers (P < 0.0001). Envelope power was elevated in the entire seizure period, and the degree of nonlinearity rose at the termination of a seizure event. Epileptic sharpness effectively characterizes an entire seizure event, complementing the role of envelope power in identifying its onset. However, the degree of nonlinearity showed superior discriminability for the termination of a seizure event. The proposed computational methods for intrinsic sharp or nonlinear EEG patterns evolving during neonatal seizure could share some features with envelope power. Current findings may be helpful in developing strategies to improve neonatal seizure monitoring.

Flexible miniature robots are expected to enter difficult-to-reach areas invivo to carry out targeted operations, attracting widespread attention. However, it is challenging for the existing soft miniature robots to substantially alter their stable shape once the structure is designed. This limitation leads to a fixed motion mode, which subsequently restricts their operating environment. In this study, we designed a biocompatible flexible miniature robot with a variable stable form that is capable of adapting to complex terrain environments through multiple movement modes. Inspired by the reversible stretching reaction of alginate saline gel stimulated by changes in environmental ion concentration, we manufactured a morphologically changeable super-soft hydrogel miniature robot body. According to the stretch and contraction shapes of the flexible hydrogel miniature robot, we designed magnetic fields for swing and rolling motion modes to realize multi-shape movement. The experimental results demonstrate that the deflection angle of the designed flexible miniature robot is reversible and can reach a maximum of 180°. The flexible miniature robot can complete forward swinging in the bar stretch state and tumbling motion in the spherical state. We anticipate that flexible hydrogel miniature robots with multiple morphologies and multimodal motion have great potential for biomedical applications in complex, unstructured, and enclosed living environments.

The conventional electrohydrodynamic (EHD) pump is limited to pumping functional and dielectric liquids, which restricts its applications in fields like microfluidics, food safety, and materials production. In this study, we present a flexible water pump driven by EHD fluid, achieved by integrating valveless elements into the fluidic channel. Our approach leverages the water-EHD interface to propel the immiscible aqueous liquid and reciprocate this process using the nozzle-diffuser system. All components of the water pump are digitally fabricated and assembled. The valveless parts are created using a laser cutting machine. Additionally, we develop a model for the EHD pump and nozzle-diffuser system to predict the generated flow rate, considering factors such as the asymmetrical performance of the EHD pump, pulse frequency, applied voltage, and structural parameters. Finally, we experimentally characterize the flow rates of both the EHD pump and water pump and apply the newly developed device to air bubble manipulation and droplet generation. This research broadens the range of specialized liquids pumped by EHD pumps to include other aqueous liquids or mixtures.

On-orbit assembly has become a crucial aspect of space operations, where the manipulator frequently and directly interacts with objects in a complex assembly process. The traditional manipulator control has limitations in adapting to diverse assembly tasks and is vulnerable to vibration, leading to assembly failure. To address this issue, we propose a human-like variable admittance control method based on the variable damping characteristics of the human arm. By collecting the velocity and contact force of human arm operations in assembly, we analyze the damping change of human arm and establish the active compliance model based on S-type damping variation rule in assembly. Furthermore, 3 passive contact models are proposed between the end of the human arm and the environment: one-sided bevel contact, both sides bevel contact, and pin-hole contact. On the basis of these active and passive models, a typical space assembly task for a robot is designed, and a human-like variable admittance controller is established and simulated. Finally, we build a ground verification platform and complete different assembly tasks, thereby successfully verifying the safety, robustness, and adaptability of the human-like variable admittance control method.

The usability of a prosthetic hand differs significantly from that of a real hand. Moreover, the complexity of manipulation increases as the number of degrees of freedom to be controlled increases, making manipulation with biological signals extremely difficult. To overcome this problem, users need to select a grasping posture that is adaptive to the object and a stable grasping method that prevents the object from falling. In previous studies, these have been left to the operating skills of the user, which is extremely difficult to achieve. In this study, we demonstrate how stable and adaptive grasping can be achieved according to the object regardless of the user's operation technique. The required grasping technique is achieved by determining the correlation between the motor output and each sensor through the interaction between the prosthetic hand and the surrounding stimuli, such as myoelectricity, sense of touch, and grasping objects. The agents of the 16-DOF robot hand were trained with the myoelectric signals of six participants, including one child with a congenital forearm deficiency. Consequently, each agent could open and close the hand in response to the myoelectric stimuli and could accomplish the object pickup task. For the tasks, the agents successfully identified grasping patterns suitable for practical and stable positioning of the objects. In addition, the agents were able to pick up the object in a similar posture regardless of the participant, suggesting that the hand was optimized by evolutionary computation to a posture that prevents the object from being dropped.

Platelet detection and counting play a greatly significant role in medical field, especially in routine blood tests which can be used to judge blood status and diagnose related diseases. Therefore, platelet detection is valuable for diagnosing related blood diseases such as liver-related diseases. Blood analyzers and visual microscope counting were widely used for platelet detection, but the experimental procedure took nearly 20 minutes and can only be performed by a professional doctor. In recent years, technological breakthroughs in artificial intelligence have made it possible to detect red blood cells through deep learning methods. However, due to the inaccessibility of platelet datasets and the small size of platelets, deep learning-based platelet detection studies are almost nonexistent. In this paper, we carried out experiments for platelet detection based on commonly used object detection models, such as Single Shot Multibox Detector (SSD), RetinaNet, Faster_rcnn, and You Only Look Once_v3 (YOLO_v3). Compared with the other three models, YOLO_v3 can detect platelets more effectively. And we proposed three ideas for improvement based on YOLO_v3. Our study demonstrated that YOLO_v3 can be adopted for platelet detection accurately and in real time. We also implemented YOLO_v3 with multiscale fusion, YOLO_v3 with anchor box clustering, and YOLO_v3 with match parameter on our self-created dataset and, respectively, achieved 1.8% higher average precision (AP), 2.38% higher AP, and 2.05% higher AP than YOLO_v3. The comprehensive experiments revealed that YOLO_v3 with the improved ideas performs better in platelet detection than YOLO_v3.

Human action representation is derived from the description of human shape and motion. The traditional unsupervised 3-dimensional (3D) human action representation learning method uses a recurrent neural network (RNN)-based autoencoder to reconstruct the input pose sequence and then takes the midlevel feature of the autoencoder as representation. Although RNN can implicitly learn a certain amount of motion information, the extracted representation mainly describes the human shape and is insufficient to describe motion information. Therefore, we first present a handcrafted motion feature called pose flow to guide the reconstruction of the autoencoder, whose midlevel feature is expected to describe motion information. The performance is limited as we observe that actions can be distinctive in either motion direction or motion norm. For example, we can distinguish "sitting down" and "standing up" from motion direction yet distinguish "running" and "jogging" from motion norm. In these cases, it is difficult to learn distinctive features from pose flow where direction and norm are mixed. To this end, we present an explicit pose decoupled flow network (PDF-E) to learn from direction and norm in a multi-task learning framework, where 1 encoder is used to generate representation and 2 decoders are used to generating direction and norm, respectively. Further, we use reconstructing the input pose sequence as an additional constraint and present a generalized PDF network (PDF-G) to learn both motion and shape information, which achieves state-of-the-art performances on large-scale and challenging 3D action recognition datasets including the NTU RGB+D 60 dataset and NTU RGB+D 120 dataset.