Medical image processing method for face dimorphism detection
Data di pubblicazione
Data di priorità
European patent application
Politecnico di Milano e Politecnico di Torino
Department of Electronics, Information and Bioengineering
Luca Bonacina, Daniele Conti, Antonio Froio, Enrico Vezzetti, Federica Marcolin
The invention is a software able to classify a population of individuals starting from 3D foetal ultrasound scans. In particular, in a first step the 3D ultrasound scans, available in the DICOM standard format, are imported, and a three-dimensional point cloud, describing the foetus’ face, is extracted by means of a double segmentation process, using a mixed 2D + 3D approach. In the second step the point clouds representing a set of faces are processed in order to identify the landmarks; the landmark detection is obtained through differential geometry techniques. Finally, the faces are classified by means of an unsupervised clustering algorithm, allowing, for instance, the detection of the group of individuals which exhibit facial dimorphisms; in the end, this kind of tool could be a great help for the physician, to diagnose diseases that would be otherwise hard to detect.
Campo di applicazione
• Pre- and post-birth medical diagnosis;
• Facial recognition;
• Phenotype analysis;
• Expression recognition.
• Standard conformity: all the file formats used during the process are standard, making the software easy to connect with other software or hardware tools already on the market;
• Fully automatic segmentation: the extraction of the 3D point cloud from the set of ultrasound slices is completely unattended, unlike most of the algorithms of this kind;
• Unsupervised clustering: the kind of classification to obtain is not needed a priori, as the clusters of individuals are generated directly by the tool;
• Geometry-based clustering: the clustering process is based on purely geometric information, computed on the surface representing the face.
Stadio di sviluppo