Optimising Neural Network Architectures for Compensator Design
| Authors: | Goodband John, Coventry University, United Kingdom
Haas Olivier, Coventry University, United Kingdom
Mills John, Walsgrave Hospital, Coventry, United Kingdom |
|---|
| Topic: | 1.1 Modelling, Identification & Signal Processing |
|---|
| Session: | Image Processing and Biomedical Applications |
|---|
| Keywords: | Compensators, Modelling, Neural Networks, Prediction Methods |
|---|
Abstract
This paper reports on investigations into optimising neural network (NN) design for predicting complex 3-dimensional compensator profiles for intensity modulated radiation therapy (IMRT) treatment. The first part of the paper describes the model used to represent compensator dimensions. The second part describes the methods used to obtain the optimal NN architecture. Results show that all three methods produce NNs capable of zero validation error using a nearest integer error criterion. The degree of accuracy obtained is within clinically accepted bounds and NNs offer a faster means for calculating compensator dimensions than existing algorithms.
