Syntactic pattern recognition-based diagnostics of fetal palates

Abstract

Analyzing ultrasound images of relatively small fetal structures is a difficult task. This difficulty particularly applies to diagnosing congenital defects of the fetus, including one of the most common defects, which is the cleft palate. To date, no methods have been developed for visualizing the fetal palate. Therefore, there is a need to improve the effectiveness of diagnostics in this area by developing appropriate image analysis and recognition methods using computer-based techniques. At the same time, relatively fast algorithms are being sought that can be a part of the software of an ultrasound device. The contribution of the paper consists in defining a new computer method satisfying all the requirements mentioned above. The method applies syntactic pattern recognition approach to the analysis of the image of the fetus’s palate. It is based on extracting of a sequence of images on multiple binarization thresholds (in the preprocessing phase), defining picture primitives on the basis of the histogram analysis, and applying a parser for a GDPLL(k) string grammar (Generalized Dynamically Programmed LL(k) grammar) for classifying abnormalities (in the recognition phase). The implementation of the method is computationally efficient and it can be a helpful tool for supporting doctors in the diagnostic process (as it has been verified in practice). The computer recognition of fetal defects on basis of the analysis of the structure of the fetus’s palate is a novel achievement. Such results have never been reported before.