Sorting 4DCT Images Based on Manifold Learning

Respiratory motion degrades anatomic position reproducibility, and result in significant errors in radiotherapy. 4D computed tomography (4DCT) can characterize anatomy motion during breathing. Usually, the acquired 4DCT images sequences is out of order. How to rearrange the sequence, i.e. sort 4DCT images has been the focus of 4DCT. In this paper we propose a method based on manifold learning, Isomap technique to reconstruct time-resolved CT volumes. By mapping high dimensional image data with Isomap into low dimensional space, each image is assigned a value, then 4DCT images is sorted according to the value to reconstruct a respiratory cycle. Experiments result shows that the method is feasible to sort 4 DCT images without using any external motion monitoring systems.