This paper proposes a graph neural network with a deformable attention mechanism to address the problems of low accuracy,large computational complexity,and poor performance in predicting vital signs in medical detection.The paper preserves all time sampled values from sensors and encodes them using a fully connected layer.The deformable attention mechanism is used as the message passing update mechanism in the graph neural network,which speeds up the prediction process.In the decoder,a multi-head attention mechanism is used for feature extraction to allow the network to observe information from multiple scales and dimensions.The input features are copied multiple times and set as separate graph nodes to enhance the adaptability of the deformable attention and the feature extraction ability of the model.Instead of a fully connected layer,a residual network is used as the decoder.In the output layer,the GeLU activation function is used instead of the traditional ReLU activation function effectively improve the accuracy of the prediction by addressing the problem of information loss on the negative half-axis.Experimental results demonstrate that the proposed model achieved high performance on three types of datasets(P19,P12,and PAM),with allperformance indicators surpassing those of other models 2.325%and higher than the best baseline performance.This indicates the effectiveness of the proposed model in predicting vital sign.