Research On The Construction Of A Dynamic Cardiac Atlas With Adjustable Shape And Motion Features

Digital heart atlas plays an essential role in cardiovascular medical image analysis,cardiac function simulation,cardiac imaging design and artificial intelligence algorithm research.Over the past decades,a variety of digital heart atlases have been developed,but most of them are limited to the modelling of ventricles.In recent years,some whole-heart atlases were proposed,but these atlases lack populational modelling of cardiac motion.In this study,a digital anatomical atlas of the whole heart incorporating heartbeat movement was constructed based on a training dataset of 57 normal human dynamic computed tomography angiography(CTA)images(each including 20 cardiac phases).The statistical shape model algorithm and linear regression method were used to construct a digital heart atlas which can change the anatomical shape and motion pattern by adjusting physiological parameters.As a pre-processing step of atlas construction,we first develop a segmentation method of the cardiac CTA image training set.The multi-atlas registration algorithm was used to segment the cardiac structures including left ventricle(LV),right ventricle(RV),left atrium(LA),right atrium(RA),and myocardium from 57 training CTA images.The resultant label image were converted into triangular meshes for the next step of atlas construction.The time-volume-curves of the ventricles and atrium were obtained based on the segmentation to provide the motion features for dynamic atlas construction.After obtaining the surface meshes of the training data,the statistical shape model algorithm was used to obtain the shape and motion variances between different people.Since the traditional statistical shape model method is only suitable for modelling three-dimensional(3D)objects,we extend this method to four dimension(4D)for the modelling of 4D(3D space+ time)objects.As a result,a 4D statistical shape model of the heart with variable shape and motions features was constructed.Based on the 4D statistical heart model,the linear regression method is further used to correlate the clinically relevant physiological parameters(e.g.,ejection fraction and systolic percentage)with the model’s shape and motion components,and finally a dynamic digital heart atlas with adjustable physiological parameters was constructed.To verify the effectiveness of the atlas,we tested the effect of physiological parameter adjustment on the changes of atlas shape and motion pattern.Changes in shape and motion pattern corresponding to changes in physiological parameters(ejection fraction and systolic percentage)were visually observed.For quantitative evaluation,the cardiac atlas was registered to the cardiac CTA images of individual patients and the accuracy of heart structure reconstruction was measured.The four heart chambers obtained an average Dice score of0.89～0.93 and an average surface distance of 1.02～1.91 mm.The digital heart atlas constructed by this institute provides a novel computational tool with adjustable shape and motion parameters for electromagnetic and biomechanical heart simulation,as well as computerized cardiac image analysis.The 4D modelling method proposed by this study is also applicable for the modelling of other moving organ structures.