The aim of this study was to measure 3D MRI images of the complete human torso which display heart and respiratory motion dynamics and to quantify image quality using different temporal discretization approaches.Biosignals (respiration belt and photoplethysmogram (PPG) sensor attached to the fingertip (BioPac MP150)) and the MR sequence trigger (at each fully sampled sphere) were recorded on a PC systems. Complex MR data were continuously acquired using a 3D radial (center out) acquisition technique with Ultra Short Echo times (UTE)1 (TE=200us, TR=1ms,FA=4°) on a clinical 3T MR system (Tim Trio, Siemens Erlangen) and multi-channel body array coils. The FoV was set to 672 mm (in each dimension) and the acquired data were reconstructed using read-out oversampling to a matrix size of 224x224x224 voxel which leads to a spatial isotropic resolution of 3 mm³. After every 50th radial spoke a navigator spoke (same radial orientation) was acquired to be able to compare external vs. intrinsic motion triggers. The total acquisition time for the 3D radial data sets was 60 minutes. 95 complete sampled k-spaces (39,520 spokes). The image reconstruction was performed using a 3D gridding algorithm with iterative sampling density estimation and optimized weighting kernel2. The biosignals were used to separate all measured radial spokes into different temporal grids. For the cardiac dynamic we used the peak-to-peak distance in PPG data and 3 different temporal resolutions (12, 15 and 20 phases). The temporal respiration information were segmented using two different approaches. The first approach separates the signal using peak-to-peak distances and the second approach uses the signal amplitude measured with the respiration belt including discrimination into in- and expiratory phases. The respiration cycle for both approaches was splitted into 8 and 10 different phases. This leads to 6 different temporal separation settings 12-8, 12-10, 15-8, 15-10, 20-8, and 20-10, where the first number indicates the number phases for a cardiac cycle and the second number for the respiration cycle, respectively. The two approaches for the respiration cycle double the number of combinations to 12. For all combinations of cardiac and respiration cycle discretization we analyzed the quality of the reconstructed data by means of SNR, CNR and evaluated under sampling artifacts by calculation of summed background signal intensities around the image object.The temporal resolution for the cardiac cycles with 12 phases was 54.25±4.17 ms, with 15 phases 43.40±3.33 ms and 32.55±2.5 ms with 20 phases, respectively. The mean duration of the respiration cycle was 5.95±1.01 s (0.17 Hz). Based on a temporal separation using peak-to-peak distances the resolution was 594.75±101.04 ms with 10 phases and 743.44±126.3 ms with 8 phases. For the phase combinations 12-8, 15-8, 15-10 and 20-10, a mean k-space coverage of 98.92%, 79.01%, 63.27% and 47.45%, respectively. If we only consider unique spokes, the mean coverage reduces to 62.56%, 54.46%, 46.77% and 37.7%, respectively. The variation of the coverage was in general below 1%. We obtained similar mean coverage values for the phase separation using signal amplitude of the respiration measure. However, the variation in k-space coverage was approx. 10 times higher including redundant and 5 times higher without considering redundant spokes. SNR and CNR correlate directly with k-space coverage, but the effect was lower as expected. SNR was found to be reduced stronger if the k-space coverage is below 50% (20-10 scheme). The graphs of Fig. 1 show the correlation of under sampling artifacts by means of background signal intensities and (unique) k-space coverage for both respiration cycle segmentation approaches.Because of the overall constant number of available spokes an increased number of specific motion phases resulted in lower k-Space coverage per motion phase, which directly affects the image quality by means of SNR and CNR. SNR and CNR start to suffer from low k-space coverage if it is below 50% while under sampling artifacts were found to increase rapidly with decreasing k-Space coverage. Furthermore, the movements of heart and respiration could be displayed more naturally and distinctive, while using an amplitude dependent assignment method for the respiratory motion phases. The main disadvantage of this method adverse to time dependent motion phases was that the specific motion phases were not equally sampled due to the influence of the breath depth and time differences between in- and expiration. Finally, we can concluded that this kind of data set offers a variety of different reconstruction approaches, e.g. displaying the heart dynamic isolated from the respiration and vice versa.