Studies have shown that heart rate variability (HRV) is a predictor of the prognosis of cardiovascular diseases. Contact heartbeat monitoring equipment is widely used, especially in hospitals, and benefits from the rapidity and accuracy of the detection of physiological health indicators. However, long‐term contact with equipment has many adverse effects. The purpose of this study was to improve the accuracy of HRV detection via noncontact equipment, thus enabling HRV to be assessed in various scenarios.

A novel deep learning approach was proposed for measuring heartbeats through camera videos. First, we performed facial segmentation and divided the face into 16 grid cells with different light balance scores. After the trend is filtered by the Hamming window, a transformer‐based neural network is used to further filter the signal. Finally, heart rate (HR) and HRV are estimated.

We used 1 million synthetic data points for pretraining and a public dataset in combination with a dataset that we constructed for task training. The final results were obtained on a test dataset that we constructed. The accuracy for HR with a low light balance score (0.867–0.983) was greater than that with a high score (0.667–0.750). Our method had higher accuracy in estimating HR than traditional filtering methods (0.167–0.417) and state‐of‐the‐art neural network filtering methods (0.783–0.917) did. The root mean square error of the HRV from the time domain was the lowest, and the correlation index score was the highest for the HRV from the frequency domain estimated by our method compared with those estimated by two neural networks.

Light balance, large sample training, and two‐stage training can improve the accuracy of HRV estimation.

Stroke is the second leading cause of death and third leading cause of disability worldwide and is the leading cause of death and disability among adults in China, with its incidence rate continuing to rise. In China, the average age of first‐time stroke patients is 66.4 years, and the intravenous thrombolysis rate using recombinant tissue plasminogen activator within 3 h of onset is only 16%. Given this fact, there is a pressing need for real‐time predictive tools, particularly for elderly individuals at home, that can provide early warnings for potential strokes.

We collected continuous monitoring data from nonintrusive smart beds and multimodal temporal data from electronic medical records at the National Center for Neurological Disorders. The data included smart bed monitoring indicators, laboratory tests, nurse observations, and static data as potential predictors, with stroke as the outcome. We applied feature representation and feature selection techniques and then input the predictors into machine learning models. Additionally, deep learning models were used after preprocessing the irregular temporal data. Finally, we evaluated the performance of the stroke prediction models and assessed the importance of the features. We used continuously updated vital signs and clinical data during hospitalization to generate timely stroke risk alerts during the same period of admission.

A total of 37, 041 samples were analyzed, of which 7020 patients were diagnosed with stroke. When only the smart bed features were used for prediction, the model achieved an area under the receiver operating characteristic curve (AUROC) of 0.59−0.63, with an accuracy ranging from 60%−65%. Among the four artificial intelligence algorithms, the random forest model demonstrated the best performance. After all the available features were incorporated, the AUROC increased to 0.94, and the accuracy improved to 92%.

In this study, the occurrence of stroke was successfully identified by integrating multimodal temporal data from electronic medical records. Noncontact monitoring of respiration and heart rate offers a promising approach for daily stroke surveillance in home‐based populations, particularly for elderly individuals living alone.