Digital Picture Processing
| Hi-index | 0.00 |
A computer support system works in our sonography laboratory providing essential services like the data-base for medical findings combined with the image data-base. The system is based on the digitization of the video-signal output of the sonograph. We have studied the possibility of easing various picture processing methods to further support the diagnosis-making process. We report here on our experience with three particular methods. So far the most successful method we have developed was the quasi-tomographical processing of images. We take up to 16 scans of the same section plane from different positions of the ultrasonic probe. The final image is obtained as a suitable average of the matched images. The method proved to be very helpful in the diagnostic process and is used on an everyday basis. On a somewhat more academic level we are studying the use of simple neural net classifiers to evaluate textural content in the images. The nets are trained on sets of texture patterns and then used to classify the testing samples. We present here two particular examples: classifying malignant tissue in testes and segmentation (sinus vs parenchyma) in kidney. Our experience with these methods is still limited. However, our conjecture is that the texture information can be used as a supportive tool in clinical praxis.