Automatisierte Erkennung und Kategorisierung der Molaren-Inzisiven-Hypomineralisation mit Hilfe künstlicher Intelligenz auf Fotografien von Zähnen

Schönewolf, Jule (2023): Automatisierte Erkennung und Kategorisierung der Molaren-Inzisiven-Hypomineralisation mit Hilfe künstlicher Intelligenz auf Fotografien von Zähnen. Dissertation, LMU München: Medizinische Fakultät

Abstract

A deep learning-based convolutional neural network (CNN) could improve dental diagnostic accuracy by automated detection and categorization of molar-incisor-hypomineralisation (MIH) on intra-oral photographs. For the purpose of this study on artificial intelligence (AI), an image set consisting of 3.241 intraoral images was split into training (N = 2.596) and test set (N = 649). The overall dataset was classified into the following categories: teeth with no signs of hypomineralisation and no dental intervention (N = 767), teeth with no signs of hypominerlisation and a MIH-related “atypical” restoration (N = 76), teeth with no signs of hypomineralisation and presence of pit and fissure sealant (N = 742), teeth with hypomineralisation and intervention (N = 815), teeth with hypominerlisation and an atypical restoration (N = 158), teeth with hypomineralisation and fissure sealing (N = 181), teeth with enamel disintegration and no intervention (n = 290), teeth with enamel disintegration and an atypical restoration (N = 169) and teeth with enamel disintegration and presence of sealing material (N = 43). After the cyclic training of the convolutional neural network most dental photographs could be automatically classified with an acceptable diagnostic accuracy. Here, an overall diagnostic accuracy of 95.2% was achieved. AUC values ranged from 0.873 (enamel breakdown with a sealant) to 0.994 (atypical restoration with no MIH). It can be concluded that AI powered MIH detection and diagnostics showed promising results in the diagnostic study. However, there is a substantial need for further improvements.

• Dokumententyp: Dissertationen (Dissertation, LMU München)
• Keywords: Chalky teeth, automated image analysis, convolutional neural network, deep learning, transfer learning
• Themengebiete: 600 Technik, Medizin, angewandte Wissenschaften ; 600 Technik, Medizin, angewandte Wissenschaften > 610 Medizin und Gesundheit
• Fakultäten: Medizinische Fakultät
• Sprache der Hochschulschrift: Deutsch
• Datum der mündlichen Prüfung: 17. Mai 2023
• 1. Bericht­erstatter:in: Kühnisch, Jan