Visual Classification of Dentoskeletal Profiles Using a Self-Organising Map (SOM)

  • Sachio Tamaoki, Section of Orthodontics, Department of Oral Growth & Development, Division of Clinical Dentistry, Fukuoka Dental College, Japan
  • Chikayo Inoue, Section of Orthodontics, Department of Oral Growth & Development, Division of Clinical Dentistry, Fukuoka Dental College, Japan
  • Sawa Nishimura, Section of Orthodontics, Department of Oral Growth & Development, Division of Clinical Dentistry, Fukuoka Dental College, Japan
  • Emiko Horikoshi, Section of Orthodontics, Department of Oral Growth & Development, Division of Clinical Dentistry, Fukuoka Dental College, Japan
  • Hiroyuki Ishikawa, Section of Orthodontics, Department of Oral Growth & Development, Division of Clinical Dentistry, Fukuoka Dental College, Japan

Objective: The purpose of this study was to create a self-organizing map (SOM) for visual classification of dentoskeletal profiles, and to demonstrate the typical patterns of profiles in malocclusion cases.
Methods: The lateral cephalometric radiographs of 109 female patients (mean age 23.9±4.1 years) with malocclusion were examined. Nineteen dentoskeletal landmarks were digitized. All points were converted into X-Y values to obtain input vectors for the SOM. By self-organization algorithm, a dentoskeletal profile map with 16 processing units was calculated. After represent inputting process of vectors, the dentoskeletal profiles of all the subjects were entered in the map.
Results: After ten thousand learning events, the dentoskeletal profile map with 4 × 4 units was calculated, producing 16 virtual profiles.
● Each virtual dentoskeletal profile was characterized by the degree of facial convexity and facial height, and antero-posterior position of the mandibule incisors.
● From the distribution of the facial profiles of the 109 subjects on the map, 5 units were found to include a relatively large number of dentoskeletal profiles.
● A virtual dentoskeletal profile of each of the 5 units was similar to the actual dentoskeletal profiles of the corresponding subjects.
Conclusion: The results suggest that this map is useful in establishing visual classification of dentoskeletal profiles based on virtual typical profiles.