An efficient medical image tracking algorithm based on motion estimation

In an ideal radiotherapy procedure, the treatment system would continuously adapt the radiation beam delivery to changes in the tumor position. The development of such a medical image-guided tracking capability is necessary. In this paper, an efficient medical image tracking algorithm based on motion estimation is proposed. The algorithm uses the motion vectors obtained from a set of selected points to calculate the parameters of the tumor motion model. It comprises three steps: the detection of feature points, the computation of correspondences between two sets of features, and the motion parameter estimation. In detail, for a pair of temporally successive pictures, feature points are extracted with the Harris detector firstly. Then, the highest-confidence-first algorithm, which first groups features for which the SAD matching error of a small window around the feature is smallest, is used in the feature matching step. The feature position gains from prediction. At last, the RANSAC parameter estimation is applied. Four random samples that are exacted in the previous step are selected and a candidate motion model is computed from these samples. All input correspondences are compared with this motion model to separate them into an inlier set and the outliers. After this, refined motion parameters are computed with a least-squares approximation on all inliers. This whole process ensures that only the motion model that had the largest number of inliers is returned as result. Experimental results show the proposed algorithm can be successfully provided with the medical image-guided tracking capability.