 |
 |
Making ultrasound images of the neonatal brain since birth is routine practice nowadays. Several things can be diagnosed from these images like asphyxia, matrix bleedings, or leukomalacia.
|
|
Advantages of ultrasound are that it is cheap, fast, and the machinery is highly portable: the ultrasound machine can be carried next to the bed of the patient, and making an ultrasound picture takes only a few seconds. This is particularly important if the patient is in need of intensive care, as is usually the case with prematurely born infants.
A disadvantage is the low image quality of ultrasound images, resulting in the necessity of an enormous amount of expertise and practice in order to be able to make any diagnosis of value. Various ways exist to pre-process the image as an aid to the subjective visual diagnosis.
The goal in our research group is to develop filtering and segmentation techniques. Filtering techniques primarily serve as a means to visually improve the image. Furthermore they can be used as a pre-processing step of a segmentation algorithm. Especially the performance of active contours can be highly improved by the removal of speckle noise.
In the figures below we give some examples of ultrasound images of the neonatal brain.
 |
 |  |
Examples of filtered images with techniques in the spatial domain are:
 |  |  |
 |
 |  |
And in the wavelet domain:
 |  |
 |
 |
 |  |
To enhance specific medical features, some techniques take into account various tissue characteristics. Changes in texture can be determined and used to determine lesions, tumors, or other affections. One of the affections of major concern in the neonatological world is "leukomalacia" (White Matter Damage). WMD is typically visible in an ultrasound image as "flares" (areas which are "whiter than normal").
 |
 |
In the scatter plot the results of measurements which distinguish the flares from healthy tissues are depicted.
Examples of filters in the spatial domain, which take into account these prior statistics to locally adapt the effect of the filter are given below. As can be seen from the examples, applying this filter as a pre-processing step for the segmentation of the flares with an active contour yields as segmentation much closer to the manual delineation of an expert than on the unprocessed image, or the images which are processed by methods that do not take into account these specific medical characteristics.
 |  |  |
 |  |  |
 |  | |
 |  |  |