基于A-CNN-LSTM模型的静脉壶凝血等级在线预测OA

Coagulation is a common complication during hemodialysis.In severe cases,it can easily lead to ir-reversible waterfall coagulation reactions.Therefore,predicting the occurrence of coagulation in advance is of great significance for ensuring the dialysis effect and life safety of patients.A coagulation grade prediction model A-CNN-LSTM(Attention-Convolutional Neural Network-Long Short-Term Memory)based on the image sequence of dialysis venous kettles is proposed to predict the coagulation grade of patients with nephropathy during dialysis online.Under the condition of ensuring the same prediction accuracy,the A-CNN-LSTM model can predict the coagulation grade for a longer time using a shorter image sequence,and the prediction accuracy of the coagulation grade for patients can reach 93.5%.Compared with the traditional prediction models LSTM and CNN-LSTM,the prediction accuracy of the proposed model has increased by 9.9 percentage points and 5.3 percentage points,respectively.After tripling the number of input frames,the prediction accuracy of the A-CNN-LSTM model reaches 94%.The experimental results show that extracting the coagulation features of venous kettles through multi-scale convolution can solve the uncertain-ty problems of the input coagulation images,the shape and size of the blood coagulation area,and also can improve the prediction accuracy.Integrating the attention mechanism into the proposed model improves the accuracy of weight distribution,accelerates the convergence speed of errors and reduces the error value.