FACIAL PHENOTYPE IDENTIFICATION FOR GENETICS DISORDER CLASSIFICATION

• Main Author: Devira Pramita, Maharani
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/48139
• Institution: Institut Teknologi Bandung

Based on data from the World Health Organization in the South East Asia Regional Office (WHO SEARO), in 2010, it was estimated that the prevalence of genetic disorders in Indonesia was 59.3 per 1000 live births. The process of getting a child's diagnosis of genetic disorders requires a series of laboratory test, while in Indonesia there are only two laboratories ready to conduct this test, namely at Cipto Mangunkusumo Hospital in Jakarta and Hasan Sadikin Hospital in Bandung, to cover Indonesia citizen from Sabang to Merauke. Seeing this, other alternatives are needed that can help the process of detecting genetic disorders. The alternative to be given is a classification model as a result of machine learning that can recognize patterns. In biology, for each rare disease due to certain genetic disorders, typically has a similar facial phenotypic. This can be utilized by machine learning algorithms to recognize patterns on faces, with facial feature extraction in the form of 68 landmark points that are spread in the face area of every human being, then calculate distance between those landmarks. The machine learning algorithm will learn the facial patterns of children in the age range of 0-12 years who have genetic disorders (which include Angelman syndrome, Cornelia de Lange syndrome, Down syndrome, and Williams syndrome) as well as those without genetic disorders. The performance evaluation of the constructed model will be assessed on the accuracy, precision, recall, and F1 matrix, based on the confusion matrix. The results of this final project are machine learning based models that can classify faces, for 2 classes with 98% accuracy (without genetic abnormalities with genetic disorders), for 3 classes with 80% accuracy (without genetic abnormalities, Down syndrome, Williams syndrome), for 4 class with 61% accuracy (without genetic disorders, Down syndrome, Williams syndrome, Angelman syndrome), and for 5 classes with 51% accuracy (without genetic disorders, Down syndrome, Williams syndrome, Angelman syndrome, Cornelia de Lange syndrome), along with phenotype characteristics that distinguish between genetic disorders and without genetic disorders.