Primena metode aktivnog modela oblikovanja u oblasti automatske identifikacije rendgenkefalometrijskih tačaka

Abstract

<p>&nbsp;&nbsp;</p>

<p>Introduction of roentgencephalometric analysis into clinical practice resulted in a revolution in the field of orthodontic diagnosis, providing precise analysis not only of dentoalveolar structures, but also of skeletal characteristics of viscero and neurocranium. There have been many attempts to further improve and automatize the roentgenocephalometric analysis in order to shorten the time for obtaining results, improve the precision and reduce errors made due to subjectivity of persons performing the analysis. Computer systems, which will automatically identify relevant skeletal and soft tissue structures and roentgencephalometric bony landmarks, will represent a great improvement and facilitate general development in the field of orthodontic diagnosis. The aim of this investigation was to examine the possibility of applying the newest generation of deformable algorithms, the so called Active Appearance Model (AAM), examining not only the structures of the contours, but also the characteristics of the complete surface structure, that is additional statistical models, which represent the pixel sensitivity variations within and around the structures. The investigation included a sample of 60 orthodontic patients treated at the Dentistry Clinic of the Faculty of Medicine in Novi Sad. All patients underwent a routine diagnostic procedure: lateral teleradiograph of the head was taken using a digital Sordex Cranex Tome Ceph, Based on the obtained images and using a specially designed software technology, 300 statistical models and textures (Active Appearance Model) were created, providing all necessary data for a complete automatic topographic and dimensional adjustment to automatic structures on anonymous images, as well as identification of relevant roentgencephalometric bony landmarks. Evaluation of AAM and of mentioned models was performed using anonymous images, whereas assessment of the radiograph quality, affecting precision of automatic detection of roentgencephalometric bony landmarks, was performed on identified images, with 70&nbsp;contrast decrease, in order&nbsp; to simulate images of lower quality. Based on the obtained results, the following conclusions were made: -It is possible to successfully perform the active appearance model in the aim of creating deformable statistical algorithms which adequately represent the average shape and variations within the structure and texture of craniofacial structures on digital teleradiographs; -The average error of automatic detection of landmarks on anonymous teleradiographs, when using Active Appearance Model, was 1.683mm for skeletal and 2.173mm for soft tissue roentgencephalometric bony landmarks; -A significant increase in detection of lowcontrast roentgenocephalometric bony landmarks was achieved in the hyperinfiltration zone with other&nbsp; craniofacial structures; - A statistically significant increase (by 50) in automatic detection of roentgencephalometric bony landmarks was achieved in regard to the related method of Active Shape Models (ASM); -By using the Active Appearance Model it is possible to detect roentgencephalometric bony landmarks with almost identical precision, even on lowquality images, where structures of interest are hardly recognizable to the human eye. Considering the established potentials and advantages of Active Appearance Model, in regard to other system, it is justly expected that by improving this method, creating precise statistical algorithms of structure and texture (based on a great number of radiographs), refining AAM algorithm in the final adaptation phase of the anonymous object, it will be possible to use it as a completely autonomous system for automatic detection of roentgencephalometric bony landmarks.&nbsp;</p>