Dose stratification-based convolutional neural networks for dose distribution prediction in radiotherapy

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate dose distribution prediction is crucial for effective radiotherapy planning.
• Current deep learning methods often overlook local dose variations, risking healthy tissue irradiation.
• There is a need for improved AI-driven dose prediction that prioritizes safety and precision.

Purpose of the Study:

• To develop a dose stratification method for refining neural network predictions in radiotherapy planning.
• To address the challenge of accurate prediction in local regions with sharp dose variations.
• To enhance the protection of healthy tissues by improving dose distribution fidelity.

Main Methods:

• A dose stratification approach is proposed, dividing dose distribution into four subcomponents.
• These subcomponents are predicted individually using neural networks in a hierarchical manner.
• A homogeneity index-based loss function is introduced to improve dose distribution uniformity.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art techniques on head and neck cancer cases.
• Experimental results on the OpenKBP dataset confirmed the method's effectiveness.
• The approach achieved dose distributions that closely align with clinically viable plans.

Conclusions:

• The dose stratification method significantly enhances the accuracy and safety of AI-based dose distribution prediction.
• This technique offers improved protection for healthy tissues by focusing on local dose variations.
• The findings increase the credibility and interpretability of artificial intelligence in radiotherapy planning.