Abstract
Wrist pulse is one kind of biomedical signals, it is affected not only by the heart beatings, but also by the conditions of nerves, organs, muscles, skin, etc. Therefore, wrist pulse signals can reflect a person’s physical state and it has been widely used in health status analysis. However, previous works mainly use traditional machine learning methods to analyze wrist pulse signal. Because wrist pulse signal is high-dimensional and complex, it is difficult for traditional machine learning methods to learn effective information from them. This study aims to explore the utilizing of deep learning methods on wrist pulse signal analysis. We propose a novel multi-kernel Convolutional Neural Network for wrist pulse signal classification. Our model can handle multiple kinds of input features and each of them will pass through a convolutional neural network that has three different sizes of convolution kernel to capture multi-scale information in different time steps. We compare our method with traditional machine learning methods on two tasks: Coronary Atherosclerotic Heart Disease Classification and Traditional Chinese Medicine Constitution yin deficiency and yang deficiency Classification. Besides, we also research the influence of different input features and different channels on wrist pulse signal analysis. The results show that our model significantly improves the performance on the two tasks, which proves the deep learning method is more suitable to deal with complex wrist pulse data.