Advanced prediction of multi-leaf collimator leaf position using artificial neural network.

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Machine Learning in Healthcare

Background:

• Multi-leaf collimators (MLCs) are essential components in modern radiotherapy, enabling precise targeting through accurate leaf positioning.
• Precise prediction of MLC leaf positions is critical for ensuring treatment efficacy and patient safety.

Purpose of the Study:

• To develop and evaluate three distinct neural network models for predicting the delivered positions of MLCs during radiotherapy treatments.
• To assess the performance of different neural network architectures in accurately forecasting MLC leaf movements.

Main Methods:

• Utilized data from 50 dynamic intensity-modulated radiation therapy plans delivered on an Elekta linear accelerator with a 160-leaf MLC system.
• Extracted features including dose fraction, gantry/collimator angles, jaw positions, and planned/adjacent leaf positions, velocity, and acceleration to train neural network models.
• Developed and compared Particle Swarm Optimization Back Propagation Neural Network (PSOBPNN), Back Propagation Neural Network (BPNN), and Radial Basis Function Neural Network (RBFNN) models, using 70% of data for training and 30% for validation/testing.

Main Results:

• All developed neural network models demonstrated high accuracy in predicting MLC leaf positions, with correlation coefficients (R) of 0.9999.
• The PSOBPNN model achieved superior performance, reporting a mean absolute error (MAE) of 0.0043 mm and a mean squared error (MSE) of 0.00003 mm².
• BPNN and RBFNN models showed higher errors, with MAE of 0.0241 mm and 0.0331 mm, respectively.

Conclusions:

• The PSOBPNN model significantly outperformed conventional BPNN and RBFNN models in predicting delivered MLC positions.
• The PSOBPNN model demonstrated robust generalizability and superior accuracy, evidenced by markedly lower MAE and MSE values.
• Accurate prediction of MLC positions using PSOBPNN enhances the safety and effectiveness of radiotherapy treatments.