Automatic projection of landmarks to generate additional correspondences in image registration

Abstract

When registering multiple multidimensional images based on landmarks, the system improves the distribution and density of the points in correspondence across images, which are of crucial importance for the accuracy and reliability of the resulting registration transform. Projection of input corresponding point landmarks is performed in order to automatically generate additional point correspondences. The input existing landmarks may have been manually or automatically located in the input multiple images. Projection is performed from each source landmark along one or more determined projection directions onto one or more determined projection targets in each image. The projection target(s) can be explicitly materialized or implicitly defined. Candidate new points are identified at the locations where the projection ray paths intersect with the projection target(s). Correspondence between the new points across images is transferred from the input landmarks from which they have been projected, with further distinction by projection direction and/or projection target in the case where a plurality of these was used. Subsequent registration can use all or a selected subset of all combined input landmark correspondences and correspondences between additional projected landmark points. The additional correspondences between projected landmark points contribute in refining the image registration. Such accurate, efficient and robust tools for image registration and any downstream processing, such as contour propagation or image fusion, are highly demanded for various medical applications, such as adaptive radiotherapy.

Description

technical field [0001] The present application is particularly applicable to medical imaging systems. However, it will be appreciated that the described techniques may also be applicable to other diagnostic systems, other imaging scenarios, or other diagnostic techniques. Background technique [0002] Segmentation of medical images is used in many applications such as radiotherapy (RT), where both tumors and organs at risk are delineated to retrospectively assess treatment delivery relative to plan and to enable anticipatory planning and adjustment of treatment. A large number of contouring tasks are performed manually or semi-manually and remain very labor intensive in clinical practice. One goal of adaptive RT is to respond retrospectively to detected deviations from the initial treatment plan, often due to changes in the patient's anatomy during the course of treatment. Such deviations can be detected from subsequent acquisitions of additional images, and can be compens...

AI Technical Summary

Problems solved by technology

For example, only landmarks on bony anatomy can be gleaned in CT data due to their reliably distinguishable or salient nature, whether by h

Images