Head radiotherapy positioning guidance system based on feature recognition and automatic annotation: Clinical...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image-Guided Therapy

Background:

• Accurate patient positioning is crucial for effective radiotherapy, particularly for head tumors due to small target volumes and proximity to critical organs.
• High precision is required to minimize damage to surrounding healthy tissues during head tumor radiotherapy.

Purpose of the Study:

• To assess the feasibility of an RGB-D camera and deep learning-based system for radiotherapy positioning guidance in head tumor localization.
• To analyze and quantify the positioning errors associated with this novel system.

Main Methods:

• Developed a system integrating deep learning algorithms (DeepLab-Opt and FFMD) with an RGB-D camera for patient surface and facial landmark detection.
• Utilized CT simulation data for creating reference contours and 3D facial landmarks, and real-time data in the radiotherapy room for comparison.
• Compared the system's positioning accuracy against the traditional cross-laser method using MVCT verification in 22 head tumor patients.

Main Results:

• The RGB-D camera and deep learning system achieved significantly lower positioning errors (lateral, longitudinal, vertical, and roll) compared to the traditional cross-laser method (p < 0.05).
• Positioning and registration time were reduced from 345.9 ± 93.4 s to 307.8 ± 36.2 s (p < 0.001).
• The system facilitated first-attempt MVCT verification, reducing patient radiation dose and improving workflow efficiency.

Conclusions:

• The developed radiotherapy positioning guidance system is feasible and effective for head tumor localization.
• It enables precise mapping between CT simulation and treatment positioning through real-time feedback on contour and facial landmarks.
• The system demonstrates significant potential for enhancing accuracy and efficiency in clinical radiotherapy applications.