Bayesian respiratory motion model : datasets

The datasets used to evaluate the Bayesian respiratory motion model are available to download.

The format of the images is MetaImage data (*.mhd).

The .mhd format is supported by Insight Toolkit (ITK) and Visualization Toolkit (VTK) libraries. In the Download section, code is provided to read/write the images using the Matlab software.

Images can be visualised and overlaid using open-source visualisation softwares such as GIMIAS and ParaView.

Database description

According to Peressutti et al, the datasets of Vol. J, Vol. K, Vol. L, and Vol. M are provided.

All imaging data of each dataset are registered to the same physical coordinate system, so images can be directly visualised and overlaid.

Each dataset consists of the following data :

Dynamic 3D MRI sequence :
- this sequence consists of 120 3D MRI images, ECG-triggered and gated at late diastole, therefore the images represent the motion of the heart due to respiration only;
- the first 40 images were acquired during normal breathing, the second 40 images during fast breathing, while the last 40 images during deep breathing;
- the field of view (FOV) covers most, but not all, of the four cardiac chambers, the typical volume size is 144 x 144 x 20, the voxel resolution is 2.43mm x 2.43mm x 3.9 mm (example of a dynamic 3D MRI image);
- these images are used to form the prior probability. For details on the pulse sequence and the prior probability formation please refer to Peressutti et al.

Number of the dynamic 3D MRI reference image :
- the progressive number of the end-exhale dynamic 3D MRI image is provided;
- this image is used as reference in the affine registration for the model formation (see Peressutti et al).

High resolution 3D MRI :
- this volume is cardiac gated at late diastole and respiratory gated at end-expiration;
- the field of view covers the whole heart and the main cardiac vessels, the typical volume size is 256 x 256 x 110, the voxel resolution is 1.37mm x 1.37mm x 1.37mm (example of a high resolution 3D MRI image);
- this high resolution image is employed to create the roadmap of the heart to be used for guidance purposes. In Peressutti et al, this volume is used for accuracy assessment and computation of the Signal-to-Noise ratio of the echo images.

Standard full-volume 3D echo images :
- 5 echo images are acquired during end-exhale breath-hold over four cardiac cycles;
- the typical volume size is 216 x 264 x 303, the voxel resolution is 0.78mm x 0.77mm x 0.68mm (example of a full-volume 3D echo image);
- the standard end-exhale echo image covering a similar FOV of the live echo images is employed as the reference image to compute the likelihood term (see Peressutti et al).

Live 3D echo images :
- 12 sequences of free-breathing 3D live echo images are provided, 3 sequences acquired during normal breathing, 3 sequences during fast breathing and 3 during deep breathing;
- 3D echo images were streamed at a rate of 14 images per second, each sequence lasted approximately 4 seconds, and thus 40-50 live 3D echo images per sequence are provided;
- the typical volume size is 199 x 94 x 211, the voxel resolution is 0.80mm x 1.04mm x 0.68mm (example of a live 3D echo sequence);
- to evaluate the Bayesian motion model, the live 3D echo sequences were manually gated at end diastole. For each sequencce, the number of the images used is specified in the file Live3D_US_gated.txt.

Surrogate signal :
- a respiratory bellows placed on the abdomen of the volunteers is used as a respiratory surrogate for the formation of the prior probability and the application of the model (see Peressutti et al);
- the file ModelFormation.txt reports 120 values of the surrogate signal, one for each dynamic 3D MRI image. These values are employed to build the affine motion model and the prior probability density function (see Peressutti et al);
- the files ModelApplication_XX.txt report the value of the surrogate signal for each live 3D echo image in the sequence number XX. These values are used as inputs to the Bayesian model in the application phase;
- positive values of the surrogate correspond to inhalation, while lower values correspond to end-exhale respiratory positions.


The following links allow to download the datasets of Vol.J, Vol. K, Vol. L and Vol. M. Each dataset contains 6 folders, one for each of the data type described in the Database description.

The MRI and the echo images are provided in a compressed (*.tar.gz) folder.

If any of the following data are used for research publication, please cite Peressutti et al.

Download dataset of Vol. J (data will be made available upon publication).

Download dataset of Vol. K (data will be made available upon publication).

Download dataset of Vol. L (data will be made available upon publication).

Download dataset of Vol. M (data will be made available upon publication).

The following Matlab scripts include an Image class for general 2D/3D image processing, functions to read/write .mhd format and a function for image transformation. Please refer to the README.txt file for script description and usage information.

Download Matlab scripts (data will be made available upon publication).


D. Peressutti, G.P. Penney, J. Housden, C. Kolbitsch, A. Gomez, E.-J.Rijkhorst, D.C. Barrat, K.S. Rhode, A.P. King. A novel Bayesian respiratory motion model to estimate and resolve uncertainty in image-guided cardiac interventions Under submission.


Devis Peressutti, Division of Imaging Sciences and Biomedical Engineering, King's College London (


Please report any problem to .