Predicting radiation therapy outcome of pituitary gland in head and neck cancer using Artificial Neural Network (ANN) and radiobiological models

Background: Pituitary dysfunction is one of the complications associated with head and neck radiation therapy. Here, radiobiological and artificial neural network (ANN) models were used to estimate the normal tissue complication probability (NTCP) of the pituitary gland. Materials and Methods: Fifty-one adult patients with nasopharyngeal carcinoma and brain tumor were studied. Two radiobiological models of Lyman Kutcher Burman (LKB), log-logistic, and ANN were employed to calculate the NTCP of the pituitary gland for all patients. BIOPLAN and MATLAB softwares were used for all calculations. The necessary parameters for each radiobiological model were calculated using Bayesian methods. R2 (coefficient of determination) and root-meansquare error (RMSE) parameters were used for the ANN method to get the best estimate. The area under the receiver operating characteristic (ROC) curve (AUC) and Akaike information criterion (AIC) were used to compare the models. Results: The respective mean NTCPs for nasopharyngeal patients with LKB and log-logistic models were 54.53% and 50.83%. For brain tumors, these values were 62.23% for LKB and 53.55% for log-logistic. Furthermore, AIC and AUC values for LKB were 77.1 and 0.826 and for log-logistic were 71.9 and 0.902, respectively. AUC value for ANN was 0.92. Conclusions: It can be deduced that LKB and log-logistic methods make reliable estimations for NTCP of the pituitary gland after radiotherapy. Moreover, the ANN approach as a novel method for NTCP calculations performed better than the two conventional analytical models as its estimations were much closer to the clinical data.