Deep Learning for Radiotherapy Outcome Prediction Using Dose Data - A Review.

Area of Science:

• Radiotherapy
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning, particularly convolutional neural networks (CNNs), is widely used for medical image analysis.
• Its application in radiotherapy prognostic modeling, especially for predicting toxicity and tumor response from dose distributions, remains limited.
• Existing studies often face challenges similar to early normal tissue complication probability models.

Purpose of the Study:

• To review and summarize studies applying deep learning to radiotherapy dose data for outcome prediction.
• To identify limitations and challenges in current deep learning models for radiotherapy.
• To highlight the potential of integrating diverse data for improved prognostic modeling.

Main Methods:

• Systematic review of studies utilizing deep learning on radiotherapy dose distributions (3D and DVH).
• Analysis of reported methodologies, data characteristics, and outcome prediction approaches.
• Identification of common issues such as small cohorts, lack of validation, and reporting deficiencies.

Main Results:

• Ten studies used spatial dose information, while four used dose-volume histograms (DVH) for prediction.
• Many studies suffer from small patient cohorts, lack of external validation, and poor reporting.
• Demonstrated technical feasibility of integrating dose, imaging, and clinical data in CNN models.

Conclusions:

• Deep learning models show potential for radiotherapy outcome prediction by integrating spatial dose information.
• Significant challenges remain, including data limitations, validation, and reporting standards.
• Further collaboration between radiation oncology and AI is needed to translate findings into clinical practice and treatment optimization.