Deformable Area-based Template Matching with Application to Low Contrast Imagery1

This paper reviews the mathematical formalism of the least squares template matching (LSM) and presents a framework for automatic quality control of the resulting match. LSM is an iterative and area-based fitting method which replaces the conventional multi-step procedure of feature extraction followed by discrete parameter search. The technique is especially suitable for attaining very high precision or for processing low-contrast, noisy and blurred imagery as it takes into account the full information of the image signal. The automatic quality control—a component often missing in commonly used image matching methods—is achieved by self-diagnostic measures supervising the iterative procedure. The development is driven by applying image analysis in order to control patient position in radiotherapy treatment. The exact positioning of patients during radiotherapy is essential for high precision treatment. Accurate information about patient position is gained by the automated matching of electronic portal images acquired during treatment sessions. However, the particular problem of such megavoltage X-ray imagery is its extremely low contrast, rendering reliable feature extraction a difficult task and thus favoring the LSM approach. The presented system includes both field edge alignment and 2D anatomy displacement measurement based on sets of template regions of stable anatomic features selected by physicians. A very promising success rate of over 90 % was achieved in clinical test data which consisted of roughly 70 image series with totally 500 portal images. Using digitally reconstructed radiographs (DRRs) as simulated portal images with known ground truth, the measurement errors were analyzed in more detail. In particular, the systematic error caused by measuring only the projection of the patient position was estimated, resulting in error bounds for the additional influence of small out-of-plane patient motion. Furthermore, promising results for the multi-modal match between a DRR reference image computed from the patient’s CT and therapeutic portal images are presented. This demonstrates the feasibility of a direct link between CT volume data with its associated treatment plan and electronic portal images, which can significantly increase the accuracy of a treatment. The implemented measuring software including a graphical user interface is running on a Workstation in the University Hospital of Zürich. This enables clinicians to carry out medical studies, which should eventually lead to an improved quality assurance in radiotherapy.