Multimodal biometrics are promising for providing a strong security level for personal authentication, yet the implementation of a multimodal biometric system for practical usage need to meet such criteria that multimodal biometric signals should be easy to acquire but not easily compromised. We developed a wearable wrist band integrated with multispectral skin photomatrix (MSP) and electrocardiogram (ECG) sensors to improve the issues of collectability, performance and circumvention of multimodal biometric authentication. The band was designed to ensure collectability by sensing both MSP and ECG easily and to achieve high authentication performance with low computation, efficient memory usage, and relatively fast response. Acquisition of MSP and ECG using contact-based sensors could also prevent remote access to personal data. Personal authentication with multimodal biometrics using the integrated wearable wrist band was evaluated in 150 subjects and resulted in 0.2% equal error rate ( ) and 100% detection probability at 1% (false acceptance rate) ( ), which is comparable to other state-of-the-art multimodal biometrics. An additional investigation with a separate MSP sensor, which enhanced contact with the skin, along with ECG reached 0.1% and 100% , showing a great potential of our in-house wearable band for practical applications. The results of this study demonstrate that our newly developed wearable wrist band may provide a reliable and easy-to-use multimodal biometric solution for personal authentication.
Early detection of arrhythmias is very important. Recently, wearable devices are being used to monitor the patient’s heartbeat to detect an arrhythmia. However, there are not satisfactory algorithms for real-time monitoring of arrhythmias in a wearable device. In this work, a novel fast and simple arrhythmia detection algorithm based on YOLO is proposed. The algorithm can detect each heartbeat on long-duration electrocardiogram (ECG) signals without R-peak detection and can classify an arrhythmia simultaneously. The model replaces the 2D ConvolutionalNeuralnetworks (CNN) with a 1D CNN and the bounding box with a bounding window to utilize raw ECG signals. Results demonstrate that the proposed algorithm has high performance in speed and mean average precisionin detecting an arrhythmia. Furthermore, the bounding window can predict different window lengths on different types of arrhythmia. Therefore, the model can choose an optimal heartbeat window length for arrhythmia classification. Since the proposed model is a compact 1D CNN model based on YOLO, it can be used in a wearable device and embedded system.
Following the COVID-19 pandemic, public interest in healthcare has significantly in-creased, emphasizing the importance of early disease detection through lung sound analysis. Lung sounds serve as a critical biomarker in the diagnosis of pulmonary diseases, and numerous deep learning-based approaches have been actively explored for this purpose. Existing lung sound classification models have demonstrated high accuracy, benefiting from recent advances in artificial intelligence (AI) technologies. However, these models often rely on transmitting data to computationally intensive servers for processing, introducing potential security risks due to the transfer of sensitive medical information over networks. To mitigate these concerns, on-device AI has garnered growing attention as a promising solution for protecting healthcare data. On-device AI enables local data processing and inference directly on the device, thereby enhancing data security compared to server-based schemes. Despite these advantages, on-device AI is inherently limited by computational constraints, while conventional models typically require substantial processing power to maintain high performance. In this study, we propose a lightweight lung sound classification model designed specifically for on-device environments. The proposed scheme extracts audio features using Mel spectrograms, chromagrams, and Mel-Frequency Cepstral Coefficients (MFCC), which are converted into image representations and stacked to form the model input. The lightweight model performs convolution operations tailored to both temporal and frequency–domain characteristics of lung sounds. Comparative experimental results demonstrate that the proposed model achieves superior inference performance while maintaining a significantly smaller model size than conventional classification schemes, making it well-suited for deployment on resource-constrained devices.