Technical Note: A feasibility study on deep learning-based radiotherapy dose calculation.

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence in Medicine

Background:

• Radiation therapy dose calculation algorithms face a critical tradeoff between computational speed and accuracy.
• Fast algorithms often lack precision, while accurate methods are time-consuming, impacting treatment planning efficiency.

Purpose of the Study:

• To explore the application of deep learning (DL) for resolving the efficiency-accuracy dilemma in radiation therapy dose calculation.
• To develop and validate a DL-based model for accurate and efficient dose computation in intensity-modulated radiation therapy (IMRT).

Main Methods:

• A modified Hierarchically Densely Connected U-net (HD U-net) deep learning model was developed for dose calculation.
• The model was trained to map initial dose distributions (from ray-tracing) to accurate dose distributions (from collapsed cone convolution/superposition).
• Feasibility was tested using prostate IMRT cases, with the model trained on 70 patients and validated on 8 separate patients.

Main Results:

• The DL model computed 3D dose distributions in approximately 1 second, demonstrating high efficiency.
• Average Gamma passing rates between DL and CS dose distributions were 98.5% (1 mm/1%) and 99.9% (2 mm/2%).
• Clinical evaluation criteria showed minimal differences (<0.25 Gy for dose, <0.16% for volume), indicating clinical equivalence.

Conclusions:

• Deep learning is feasible for accurate and efficient radiotherapy dose calculation.
• The developed DL model offers a promising solution for optimizing radiation therapy treatment planning.