Tumor location displacement caused by respiration-induced motion reduces the efficacy of radiation therapy. Three medically relevant patterns are often observed in the respiration-induced motion signal: baseline shift, ES-Range shift , and D-Range shift. In this paper, for patients with lower body cancer, we develop class profiles (a low dimensional pattern frequency structure) that characterize them in terms of these three medically relevant patterns. We propose an adaptive segmentation technique that turns each respiration-induced motion signal into a multi-set of segments based on persistent variations within the signal. These multi-sets of segments is then probed for base behaviors. These base behaviors are then used to develop the group/class profiles using a modified version of the clustering technique described in [1]. Finally, via quantitative analysis, we provide a medical characterization for the class profiles, which can be used to explore breathing intervention technique. We show that, with i) carefully designed feature sets, ii) the proposed adaptive segmentation technique, iii) the reasonable modifications to an existing clustering algorithm for multi-sets, and iv) the proposed medical characterization methodology, it is possible to reduce the time series respiration-induced motion signals into a compact class profile. One of our co-authors is a medical physician and we used his expert opinion to verify the results.

Accurate motion estimation in thoracic computed tomography (CT) plays a crucial role in the diagnosis and treatment planning of lung cancer. This paper provides two key contributions to this motion estimation. First, we show we can effectively transform a CT image of effective linear attenuation coefficients to act as a density, i.e. exhibiting conservation of mass while undergoing a deformation. Second, we propose a method for diffeomorphic density registration for thoracic CT images. This algorithm uses the appropriate density action of the diffeomorphism group while offering a weighted penalty on local tissue compressibility. This algorithm appropriately models highly compressible areas of the body (such as the lungs) and incompressible areas (such as surrounding soft tissue and bones).