Comparative Analysis of Neural Network Training Methods in Real-time Radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Artificial Intelligence in Medicine

Background:

• Patient motion during radiotherapy, especially in the chest, poses a significant challenge to protecting healthy tissues from high radiation doses.
• Real-time radiotherapy techniques utilize markers to track patient movement, enhancing dose delivery accuracy to the tumor.
• Accurate tumor targeting is essential for effective cancer treatment and minimizing side effects.

Purpose of the Study:

• To evaluate the accuracy of artificial intelligence (AI) methods for estimating tumor positions in real-time radiotherapy.
• To compare the performance of neural networks, genetic algorithms, and particle swarm optimization (PSO) in real-time tumor tracking.
• To assess the potential of AI in improving the precision of radiotherapy delivery.

Main Methods:

• Utilized recorded signals from three external markers over 10 breathing cycles of a lung cancer patient undergoing cyber-knife treatment.
• Applied neural network, genetic algorithm, and particle swarm optimization (PSO) methods using MATLAB software.
• Trained and tested AI models to determine tumor locations based on marker signal data.

Main Results:

• Neural network achieved a high accuracy of 0.8% in estimating tumor positions.
• Genetic algorithm and particle swarm optimization (PSO) showed lower accuracies of 12% and 14%, respectively.
• Demonstrated the superior performance of the neural network method for real-time tumor localization.

Conclusions:

• Neural network algorithms are highly effective for determining internal target volume (ITV) during radiotherapy.
• AI-driven tumor position estimation enhances the precision of real-time radiotherapy delivery.
• The findings support the integration of advanced AI techniques in radiation oncology for improved patient outcomes.