Linear Regression Analysis of 2 D Projection Image Data of 6 Degrees - of - Freedom ArcHNES Transformed 3 D Image Sets for Stereotactic Radiation Therapy

Patient positioning is crucial to accurate dose delivery during radiation therapy to ensure the proper localization of dose to the target tumor volume. In patient positioning for stereotactic radiation therapy treatment, classical image registration methods are computationally costly and imprecise. We developed an automatic, fast, and robust 2D-3D registration method to improve accuracy and speed of identifying 6 degrees-of-freedom (DoF) transformations during patient positioning for stereotactic radiotherapy by creating a model of characteristic shape distributions to determine the linear relationship between two real-time orthogonal 2D projection images and the 3D volume image. We defined a preprocessed sparse base set of shape distributions that characterize 2D digitally reconstructed radiograph (DRR) images from a range of independent transformations of the volume. The algorithm calculates the 6-DoF transformation of the patient based upon two orthogonal real-time 2D images by correlating the images against the base set The algorithm has positioning accuracy to at least 1 pixel, equivalent to 0.5098 mm accuracy given this image resolution. The shape distribution of each 2D image is created in MATLAB in an average of 0.017 s. The online algorithm allows for rapid and accurate position matching of the images, providing the transformation needed to align the patient on average in 0.5276 s. The shape distribution algorithm affords speed, robustness, and accuracy of patient positioning during stereotactic radiotherapy treatment for small-order 6-DoF transformations as compared with existing techniques for the quantification of patient setup where both linear and rotational deviations occur. This algorithm also indicates the potential for rapid, high precision patient positioning from the interpolation and extrapolation of the linear relationships based upon shape distributions.