Predicting Total Dwell Time of IPSA Plan Based on Machine Learning and Dose Calculation Models

Objective To establish two models based on machine learning and dose calculation algorithm that can be used for the prediction of the total dwell time and rapid quality control of brachytherapy plans.Methods A total of 1042 cases of treated gynecologic oncology patients were selected,of which 512 were used as training data to establish the model and the rest were used as test data.Each treatment plan optimized by inverse planning simulated annealing with all three catheters of the Fletcher applicator.The source strength S,prescription dose D,source dwell time t,and tumor volume V were recorded for each case.Rwas defined as S·t/D.In accordance with the prescription dosage calculation formula in the planning system and machine learning method,the following equations were established:R=kVand R=a+b · V+c·V~2.The R~2 correlation coefficient represents the accuracy of the results.Result The dose calculation algorithm-based model is R=1272×V,R~2=0.959,whereas the machine learning-based model is R=258.8× V-0.359× V~2+5110,R~2=0.961.The treatment time prediction of the two models,each having 13 and 15 cases,respectively,has an error rate of more than10%,and the dose calculation algorithm-based method is more accurate.Conclusion The treatment time can be quickly predicted according to the planning target volume,and the two prediction models can be used as a way of quality control.