Prediction of Multileaf Collimator Leaf Positions Using a Hybrid BKA– Neural Network Model for Radiotherapy Delivery Accuracy

Abstract Methods The Black‑winged Kite Algorithm (BKA) was integrated with a three-layer BP Neural Network (BKA-BPNN), where BKA optimized the initial weights and thresholds. Thirteen dosimetric and mechanical parameters were extracted from the treatment planning log files as inputs to the model, with the delivered leaf positions serving as the target response. The model’s performance was evaluated using various metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), regression plots, and error histograms. Results The improved neural network model (BKA-BPNN) accurately predicted MLC leaf positions, demonstrating superior performance (MAE = 0.0026 mm, MSE = 0.00001 mm²). A correlation coefficient of 0.9999 indicates a high correlation between the predicted and actual leaf positions. Additionally, the prediction efficiency has significantly improved. Compared with the traditional BPNN model, the BKA-BPNN model reduces the MAE by 89.21%. When compared with the improved Particle Swarm Optimization Back Propagation Neural Network (PSO-BPNN) model, it achieves a 39.53% reduction in MAE. Conclusion The proposed BKA-BPNN model outperforms both conventional neural networks and deep learning-based neural networks in prediction accuracy, which possesses great application potential for boosting the precision of radiotherapy treatment.