Respiratory Motion Modeling Based On4D Lung CT Image

In recent years, as medical imaging has recently undergone rapid development,the radiation therapy or interventional operation to navigate using image informationhas been widely welcomed. At present, the clinical image guided system utilizes thestatic image in the case of static organ positions and shapes. For the lung and liver,the deformation of these organs are deformable, their positions and shapes willchange obviously due to breathing. The image guided system using the static imagewill cause inaccurate lesion localization, so we wish to construct a respiratorymotion model and realize dynamic image-guided diagnosis and treatment.To construct the respiratory motion model, first, it needs to use the4D lung CTimage acquired over several respiratory cycles and integrate the structuralinformation of diaphragm, tracheas and vessels; subsequently, deformableregistration is performed to obtain the deformation fields within the same respiratorycycle at different phases; finally, functional relationship between respiratory signaland deformation displacements is established in combination with an external signal.However, the existing methods of building the motion models have their ownlimitations in process of activity implementation.Aiming at the disadvantages of time-consuming and great dose radiation inacquiring the4D lung CT image, and the inaccuracy of building the sequentialcorrespondence between4D lung CT image and the measured3D coordinates of the labeling, a novel method for constructing a patient specific respiratory motion modelusing a limited number of4D lung CT image was proposed in this paper. First, the4D lung CT image that were used in the study were acquired on a Toshiba AquilionONE320-slice CT scanner using the ’Lung mass perfusion’ protocol while the3Dcoordinates of in vitro labeling were measured as an external respiratory signal, theCT scanner can implement dynamic imaging of the whole-lung under free breathing.Furthermore, a distance correlation method was used to build the sequentialcorrespondence between the4D lung CT image and external respiratory signal.Subsequently,4D lung CT deformable registration was performed to acquire the3Ddisplacements for every registration control point with respect to time. The temporalfitting was adopted to obtain both the external respiratory signal fitting curve and thecontrol point displacements fitting curves. Finally, the motion model was constructedby fitting a linear function that related each control point displacement to thecorresponding respiratory signal value. Three patient datasets were used in this study;the model error was calculated to quantitatively evaluate the modeling’s accuracy.The results show that the proposed method is successful in striking a balancebetween modeling accuracy and number of4D lung CT image, and therefore, it is apotentially useful tool for constructing a lung respiratory motion model with a lowdose that will aid in accurately locating a lesion in surgery path planning and surgicalnavigation.