[Automatic detection of perfusion deficits with Bolus Harmonic Imaging]

The diagnosis of ischemic stroke relies increasingly on the usage of ultrasound-based methods. One of the recent methods is the transcranial, contrast agent-based Bolus Harmonic Imaging (BHI) method. The captured image sequence is manually examined by clinical experts thus resulting in a time-consuming procedure. The purpose of this study is to evaluate three different methods to analyze BHI image sequences automatically for the detection of ischemic brain tissue.

BHI captures an image sequence that provides information on the dynamic behavior of the ultrasound contrast agents. This image sequence is analyzed using three different procedures. First a system relying on expert knowledge is used to determine perfusion defects. This procedure requires parametric images, which are previously extracted from the image sequence. The parameter images are then categorized by an unsupervised classification method in well-perfused and ischemic tissue by regarding the parametric images as features describing the perfusion. Thirdly, the whole image sequence can be interpreted as a pixel-by-pixel behavior out of contrast agents. The dynamic curve of each pixel can be automatically classified as perfused and ischemic tissue by the K-Means method without extracting parametric images. In all three cases a closing step is necessary for the accurate interpretation of the results. Transcranial ultrasound imaging produces typical stripe artifacts that have to be detected and eliminated. A result image is then created and provides a conclusion about perfusion reduction in brain tissue.

All three methods have been validated on the basis of 26 patients by clinical experts. The segmentation on the contrast agent kinetics has proven to be most effective. According to our patient database, it provides the highest detection accuracy, resulting in a sensitivity of 100% and a specificity of 100%.

The presented methods seem to be adequate for detecting ischemic brain tissue. The classification of contrast agent kinetics provides the best results and has further advantages. It is robust with respect to noise and the calculation is fast because the extraction of parametric images is omitted. The very high sensitivity and specificity must be validated in a larger patient population. Reliable and automated detection of perfusion defects at the bedside seems to be possible.