Improved Quantification of the Heart by Utilizing Images from Different Imaging Directions

by Jyrki Lötjönen, Juha Koikkalainen and Kirsi Lauerma

Since cardiovascular disease is the most common cause of death in the Western countries, there is a strong need to diagnose and to treat cardiac diseases. The cardiac imaging techniques, such as ultrasound and magnetic resonance imaging (MRI), have improved considerably during the last years providing nowadays detailed anatomical and functional information on the heart. We have developed a novel method for extracting quantitative measures of the heart from differently oriented image directions.

In clinical practice, the interpretation of the images is still often performed visually due to lack of automatic tools for extracting quantitative measures from images. Although visual inspection of images provides in many cases enough signs for the interpretation of images, more objective and accurate quantitative measures are needed, for example, for detecting subtle indicators related to early phases of diseases or comparison of different populations in drug-discovery studies.

Although MRI can be regarded as a golden standard for extracting three-dimensional (3D) anatomical and functional information on the heart, commercially available products provide tools only for the analysis of the left ventricle, and the tools require often a substantial amount of manual interaction. In addition, the quantitative analysis is often performed only for short-axis images (see figure). The slice thickness is usually several times higher than the resolution in the image plane, especially in functional cardiac images. For this reason, it is difficult to localize accurately different structures on the thickness direction, such as the apex and the valve levels from short-axis images. Long-axis images, which are orthogonal to the short-axis images, are often acquired in clinics providing accurate data on the problematic direction in the short-axis images.

We have developed a technique where image series from two or more imaging directions are used for extracting quantitative measures from images. In this work, the technique was applied for computing volumetric measures from the ventricles and atria of the heart. The procedure consists of the following steps:

• Define the spatial relationship between the images from different imaging directions. The location of each pixel in the co-ordinate system of an imaging device can obtained from image headers. However, breathing causes movement artefacts (even more than 2 cm) which need to be corrected separately. We have developed a technique where the location of each short-axis image is optimized based on the information on long-axis images and vice versa.
• Segment the structures from images by registering non-rigidly (warping) an a priori model (template) to all image series simultaneously. Our model was built from a database of 25 healthy volunteers; the model represents the mean shape and appearance of a healthy human heart (see figure). The similarity between the model and the data to be segmented was maximized. In addition, the typical variation of the shape was also modelled from the database and used to constrain the deformation.
• Compute the volumes of the objects of interest.

Model: The mean model consists of short- and long-axis image series computed from 25 healthy subjects and a 3D surface model of atria, ventricles and epicardium. The white curves on the images show the surface model superimposed on the mean grey-scale images. Result: Segmented short- and long-axis images and the volume of atria and ventricles from one subject during a cardiac cycle.

The figure represents segmentation results of one subject using the technique described above. The volumes of left ventricle (LV), right ventricle (RV), left atrium (LA) and right atrium (RA) at different phases of the cardiac cycle are also represented. Two major improvements can be emphasized:

1) The automatic definition of the volumes of atria has not been reported earlier. The volumetry of atria is interesting because it is known that the symptoms in some diseases appear first (at early phase) in atria and later in ventricles.

2) The volumes of ventricles are defined more accurately than earlier. Up to 30 % (average 10-15%) differences were detected in the volumes of ventricles, as the results using only short-axis images and using both short- and long-axis images were compared.

In the near future, the software tool will be installed into clinical environment and the user interface will be modified to fulfil the requirements of medical experts.

Please contact:
Jyrki Lötjönen, VTT Information Technology, Finland.
Tel: +358 3 316 3378
E-mail: jyrki.lotjonenvtt.fi

Juha Koikkalainen,
Helsinki University of Technology, Finland.
E-mail: juha.koikkalainenhut.fi

Kirsi Lauerma,
Helsinki University Central Hospital, Finland.
E-mail: kirsi.lauermahus.fi