Recognition of dysmorphic syndromes using image analysis

The lexial meaning of the dysmorphic is defined as the congenital malformation ofhuman. At the present time, the most common syndrome of the dysmorphicautosomal chromosome diseases is Trisomy 21, in other words down syndrome.Therefore, clinic pre-diagnosis of down syndrome carries severity. Clinic prediagnosiscan be estimated by either comparison of the images on reference booksor experience which can show difference from one clinician to other. On this study, inorder to obstruct the dissimilarity of prediagnosis for the patients who are douptedlikely down syndrome and to render this process clinican independent, it is aimed todetermine the clinic prediagnosis by the image analysis subsequently qualitativelyobservation of the comparison method. Regarding our study, a database has beenconstituted with the face photos of 18 children who has already been diagnoseddown syndrome and 18 children who has normal morphology. At the MATLAB basedprogram which is written for our thesis, the fiducial points on faces are determined byusing the elastic face bunch graph method for all photos. Afterwards, 10 featurevectors for all faces are obtained from these fiducial points for cilinic prediagnosis.Feature vectors are used for training the program by artificial neural networks. Inconclusion, by using two different artificial neural network method, the determinationof clinic prediagnosis for a patient who has down syndrome can be done with anaccuracy of 68%. For further studies, it will be possible to increase the success ratioby creating larger databases. As a result of these studies, we will be able to reach astandardiazation for pre-diagnosis of dysmorphic diseases.