Stroke is a world-leading disease for causing disability. Brain-computer interaction (BCI) training has been proved to be a promising method in facilitating motor recovery. However, due to differences in each patient’s neural-clinical profile, the potential of recovery for different patients can vary significantly by conducting BCI training, which remains a major problem in clinical rehabilitation practice. To address this issue, the objective of this study is to prognosticate the outcome of BCI training using motor state electroencephalographic (EEG) collected during the first session of BCI tasks, with the aim of prescribing BCI training accordingly. A Convolution Neural Network (CNN) based prognosis model was developed to predict the outcome of 11 stroke patients’ recovery following a 2-week rehabilitation training with BCI. In our study, functional connectivity and power spectrum have been evaluated and applied as the inputs of CNN to regress patients’ recovery rate. A saliency map was used to identify the correlation between EEG channels with the recovery outcome. The performance of our model was assessed using the leave-one-out cross-validation. Overall, the proposed model predicted patients’ recovery with R ² 0.98 and MSE 0.89. According to the saliency map, the highest functional connectivity occurred in Fp2/Fpz-AF8, Fp2/F4/F8-P3, P1/PO7-PO5 and AF3-AF4. Our results demonstrated that deep learning method has the potential to predict the recovery rate of BCI training, which contributes to guiding individualized prescription in the early stage of clinical rehabilitation.

Since the underlying mechanisms of neurorehabilitation are not fully understood, the prognosis of stroke recovery faces significant difficulties. Recovery outcomes can vary when undergoing different treatments; however, few models have been developed to predict patient outcomes toward multiple treatments. In this study, we aimed to investigate the potential of predicting a treatment's outcome using a deep learning prognosis model developed for another treatment. A total of 15 stroke survivors were recruited in this study, and their clinical and physiological data were measured before and after the treatment (clinical measurement, biomechanical measurement, and electroencephalography (EEG) measurement). Multiple biomarkers and clinical scale scores of patients who had completed manual stretching rehabilitation training were analyzed. Data were used to train deep learning prognosis models, yielding an 87.50% prognosis accuracy. Pre-trained prognosis models were then applied to patients who completed robotic-assisted stretching training, yielding a prognosis accuracy of 91.84%. Interpretation of the deep learning models revealed several key factors influencing patients' recoveries, including the plantar-flexor active range of movement (r = 0.930, P = 0.02), dorsiflexor strength (r = 0.932, P = 0.002), plantar-flexor strength (r = 0.930, P = 0.002), EEG power spectrum density and EEG functional connectivities in the occipital, central parietal, and parietal areas. Our results suggest (i) that deep learning can be a promising method for accurate prediction of the recovery potential of stroke patients in clinical scenarios and (ii) that it can be successfully applied to different rehabilitation trainings with explainable factors.

Various rehabilitation robots have been developed to assist the movement training of the upper limbs of stroke patients, among which some have been used to evaluate the motor recovery. However, how to understand the recovery of motor function from the quantitative assessment following robot-assisted rehabilitation training is still not clear. The objective of this study is to propose a quantitative assessment method of motor function based on the force and trajectory characteristics during robotic training to reflect motor functional recovery. To assist stroke patients who are not able to move voluntarily, an assistive training mode was developed for the robot-assisted rehabilitation system based on admittance control. Then, to validate the relationship between characteristic information and functional recovery, a clinical experiment was conducted, in which nine stroke patients and nine healthy subjects were recruited. The results showed a significant difference in movement range and movement smoothness during trajectory tracking tasks between stroke patients and healthy subjects. The two parameters above have a correlation with the Fugl-Meyer Assessment for Upper Extremity (FMU) of the involved patients. The multiple linear regression analysis showed FMU was positively correlated with parameters ($R^2=0.91,p<0.005$). This finding indicated that the above-mentioned method can achieve quantitative assessment of motor function for stroke patients during robot-assisted rehabilitation training, which can contribute to promoting rehabilitation robots in clinical practice.

Predicting the potential for recovery of motor function in stroke patients who undergo specific rehabilitation treatments is an important and major challenge. Recently, electroencephalography (EEG) has shown potential in helping to determine the relationship between cortical neural activity and motor recovery. EEG recorded in different states could more accurately predict motor recovery than single-state recordings. Here, we design a multi-state (combining eyes closed, EC, and eyes open, EO) fusion neural network for predicting the motor recovery of patients with stroke after EEG-brain-computer-interface (BCI) rehabilitation training and use an explainable deep learning method to identify the most important features of EEG power spectral density and functional connectivity contributing to prediction. The prediction accuracy of the multi-states fusion network was 82%, significantly improved compared with a single-state model. The neural network explanation result demonstrated the important region and frequency oscillation bands. Specifically, in those two states, power spectral density and functional connectivity were shown as the regions and bands related to motor recovery in frontal, central, and occipital. Moreover, the motor recovery relation in bands, the power spectrum density shows the bands at delta and alpha bands. The functional connectivity shows the delta, theta, and alpha bands in the EC state; delta, theta, and beta mid at the EO state are related to motor recovery. Multi-state fusion neural networks, which combine multiple states of EEG signals into a single network, can increase the accuracy of predicting motor recovery after BCI training, and reveal the underlying mechanisms of motor recovery in brain activity.

INTRODUCTION: Reliable early detection of Alzheimer's disease (AD) remains difficult due to heterogeneous progression trajectories and variability in clinical presentation. Multimodal approaches leveraging neuroimaging and clinical data offer promise but often struggle with effective integration and generalization. METHODS: We developed Alzheimer's Disease Language and Image Pre‐Training (ADLIP), a vision–language framework that integrates 3D T1‐weighted magnetic resonance imaging with structured clinical records. The model uses a multi‐teacher training strategy to enhance generalizability and robustness across modalities, enabling more reliable representation learning for AD diagnosis. RESULTS: ADLIP outperformed baseline CLIP and fine‐tuned CLIP models in three‐class classification and zero‐shot diagnosis, achieving improved accuracy, F ₁ score, and macro‐area under the curve. Longitudinal evaluation demonstrated alignment with clinical trajectories. ADLIP maintained performance across racial subgroups without retraining. DISCUSSION: Our results demonstrate that contrastive multimodal representation learning enables clinically meaningful, generalizable, and temporally stable AD diagnosis across diverse populations. Highlights: A novel vision–language foundation model (Alzheimer's Disease Language and Image Pre‐Training [ADLIP]) integrates 3D magnetic resonance imaging and clinical text for Alzheimer's disease diagnosis. ADLIP enables zero‐shot prediction of unseen data and cognitive scores without task‐specific fine‐tuning. The model demonstrates strong generalizability across racially diverse cohorts, supporting equitable clinical use. Longitudinal evaluation shows alignment with disease progression, highlighting utility for monitoring applications.

Functional corticomuscular coupling (FCMC) probes multi-level information communication in the sensorimotor system. The canonical Coherence (caCOH) method has been leveraged to measure the FCMC between two multivariate signals within the single scale. In this paper, we propose the concept of multiscale canonical Coherence (MS-caCOH) to disentangle complex multi-layer information and extract coupling features in multivariate spaces from multiple scales. Then, we verified the reliability and effectiveness of MS-caCOH on two types of data sets, i.e., a synthetic multivariate data set and a real-world multivariate data set. Our simulation results showed that compared with caCOH, MS-caCOH enhanced coupling detection and achieved lower pattern recovery errors at multiple frequency scales. Further analysis on experimental data demonstrated that the proposed MS-caCOH method could also capture detailed multiscale spatial-frequency characteristics. This study leverages the multiscale analysis framework and multivariate method to give a new insight into corticomuscular coupling analysis.