Prediction of normal tissue complication probability and equivalent uniform dose of organs at risk in lung cancer treatment plans using an Artificial Neural Network

The purpose is to predict Normal Tissue Complication Probability (NTCP) and Equivalent Uniform Dose (EUD) in external beam radiation therapy of lung cancer treatment plans by applying an Artificial Neural Network (ANN). 100 lung cancer treatment plans of anonymized patients were selected. NTCP of Organs At Risk (OARs) and EUD were calculated for all plans using an analytical EUD-based linear quadratic model. The calculated EUD model-based NTCP and EUD were used as outputs for training and testing. The inputs for the ANN model were treatment modality, location of tumor, prescribed dose, fractions, planning target volume (PTV), number of fields, mean dose to PTV, gender, age, and mean doses to the OARs. All numeric data were normalized in the range of 0 to 1, while categorical inputs were changed to 0 or 1. Our ANN is based on Levenberg-Marquardt back-propagation algorithm with one hidden layer having 16 nodes with 13 inputs and 2 outputs. 70% of the data were employed for training, 15% for validation, and 15% for testing. The ANN performance was evaluated by mean squared error (MSE) and correlation coefficients (R) of regression. In NTCP and EUD prediction, the average correlation coefficients are 0.94 for training, 0.89 for validation, and 0.84 for testing. The maximum ANN mean squared error (MSE) is 0.025 in predicting NTCP and EUD of the heart. These results indicate that our ANN can be employed to predict NTCP and EUD.