Digital portal image registration by sequential anatomical matchpoint and image correlations for real-time continuous...

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Analysis

Background:

• Accurate radiotherapy requires precise alignment of radiation fields using electronic portal imaging devices (EPIDs).
• Real-time detection of field misalignments necessitates rapid image registration (under 250 ms).
• Current registration methods are sensitive to user accuracy in identifying anatomical landmarks.

Purpose of the Study:

• To develop an improved sequential digital portal image registration method.
• To enhance the accuracy of radiotherapy field alignment and misalignment measurement.
• To enable rapid, real-time detection of field placement errors.

Main Methods:

• A two-pass registration algorithm combining user-identified anatomical matchpoints and image information.
• First pass: Generates initial parameters accounting for matchpoint uncertainty.
• Second pass: Maximizes cross-correlations for improved registration quality, avoiding intensive computations.

Main Results:

• Registration achieved in under 6 ms per matchpoint on a 486 CPU.
• Errors: <2% magnification, <0.5° in-plane rotation, <1 pixel in-plane translation.
• The algorithm is sensitive to out-of-plane rotations due to rigid-body geometry assumption.

Conclusions:

• The developed sequential registration method is fast and accurate for real-time EPID applications.
• It offers significant improvements over methods relying solely on user-identified landmarks.
• Further analysis quantifies accuracy and highlights sensitivity to specific rotational errors.