An Artificial Intelligence-Based Full-Process Solution for Radiotherapy: A Proof of Concept Study on Rectal Cancer

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Rectal cancer radiotherapy requires complex and time-consuming treatment planning.
• Current manual processes are prone to variability and inefficiency.

Purpose of the Study:

• To develop and evaluate an artificial intelligence-based, full-process solution for rectal cancer radiotherapy.
• To integrate autosegmentation and automatic treatment planning within a single deep-learning framework.

Main Methods:

• A deep learning framework utilizing a convolutional neural network (CNN) for segmentation and dose distribution.
• Plan optimization executed via a Pinnacle script simulating the treatment planning process.
• Model training and validation on 172 and 18 rectal cancer patients, respectively; end-to-end evaluation on 40 patients.

Main Results:

• Full-process planning achieved in 7 minutes, with an additional 15 minutes for contour modification and re-optimization.
• High segmentation accuracy for Planning Target Volume (PTV) and Organs at Risk (OARs) with DICE similarity coefficients > 0.85.
• Physician acceptance rate of 80% for auto-generated plans without further modification, indicating comparable quality to manual plans.

Conclusions:

• A novel deep learning-based automatic solution for rectal cancer radiotherapy planning has been successfully developed.
• This AI-driven approach significantly enhances the efficiency of the treatment planning process.