Application of deep learning in radiation therapy dose calculation: A systematic review

Radiation therapy dose calculation is an essential component of cancer treatment. Radiation therapy outcome depends on the accurate calculation and delivery of the prescribed dose to the tumor while sparing the surrounding healthy tissues. Commercially available dose calculation methods are based on mathematical models that approximate the interactions of radiation with human tissue. However, they are not perfect and can sometimes produce inaccurate results depending on the treatment planner's skills. Moreover, these traditional dose calculation algorithms are often time-consuming. Regarding the increasing application of artificial intelligence and deep learning methods in medical sciences, scientists have used these methods to overcome the aforementioned issues in the treatment planning of cancer patients. Therefore, this study aims to review the application of deep learning methods in radiation therapy treatment planning. PubMed, Science Direct, Web of Science, and Google Scholar databases were explored up to May ۲۰۲۲, using different combinations of the keywords: "radiation therapy", "treatment planning", "artificial intelligence", " dose calculation ", and "deep learning". After screening the results ۱۰ more recent and relevant papers were included in the study. Deep learning models require a combination of patient anatomy data, linear-accelerator intensity modulated radiation therapy (IMRT) multi-leaf-collimator shape or segment data, dose data, and physics-based inputs to accurately predict the dose distribution of the radiation beam in the patient's body, for radiotherapy dose calculation. By training a deep learning model on a large dataset of radiation therapy plans, it is possible to learn the complex relationships between input and output features. This allows deep learning models to capture subtle patterns in the data that may be missed by traditional mathematical models. The dose prediction using the trained network is very fast, making it compelling for online adaptive workflows where fast segment dose calculations are needed. Deep learning methods, such as deep U-Net algorithm have shown promising results in treatment planning by providing fast and accurate dose calculation. Additionally, deep learning algorithms can boost the accuracy of less accurate dose calculation algorithms by capturing the differences between dose calculation algorithms. In conclusion, deep learning-based methods provide a more efficient and accurate alternative to traditional and commercially available dose calculation methods.

کلیدواژه ها:

Artificial Intelligence ، Deep Learning ، Radiation Therapy ، Dose Calculation ، Treatment Planning.