摘要
Abstract
Objective To develop a precise model for predicting parotid gland dose in radiotherapy,so as to achieve personalized tumor treatment and improve the therapeutic effect.Methods Long short-term memory(LSTM)neural networks in combination with the sparrow search algorithm(SSA)were employed for parameter optimization to construct a parotid gland dose prediction model.Various data factors relevant to dose prediction were collected,and the model against others was compared and analyzed to validate its accuracy and prediction errors.Results The SSA-LSTM model showed higher accuracy and stability in dose prediction of parotid D15,D30,D45 and Dmean.When predicting the parotid D15 test set,the mean absolute error(MAE)of SSA-LSTM was 0.2966,and the goodness of fit R2 was 0.9663.Relatived to LSTM,LSTM optimization based on genetic algorithm,LSTM utilized grey wolf optimization algorithm,the MAE reduction rates of SSA-LSTM were 40.93%,33.39%and 25.51%,respectively,and the R2 increase rates were 8.06%,4.49%and 3.03%,respectively,which proved the superiority of SSA-LSTM model compared with other optimization algorithms in parotid gland dose prediction in radiotherapy.The reliability and stability of SSA-LSTM model were also verified by the analysis of Taylor diagram.Conclusion The SSA-optimized LSTM model can significantly improve the accuracy of parotid gland dose prediction.Moreover,This model can be extended to other areas of radiation therapy and have a positive social significance in the medical field.关键词
放射治疗/个体化剂量预测/腮腺剂量/麻雀搜索算法/长短期记忆神经网络Key words
radiotherapy/individualized dose prediction/parotid gland dose/sparrow search algorithm/long short-term memory neural network分类
医药卫生
