Skip to main content
ICP TPID JOINT MEETING

Papers Proceedings »

Developing A Deep Learning Network for Recognition of Dental Implant Systems

Since implants began to be used in clinical practice, more than 50 years have passed, various complications have been reported with implants. To manage such complications, it is necessary to have various information such as the manufacturer of the implant, its system (tissue level, bone level, placement in cortical or cancellous bone), diameter, and abutment type. If there is no record of which implant was used, it is a very important factor to accurately define the current implant brand and system only from the patient's radiography, in order to manage complications or remove the implant. Unfortunately, there are very few studies on methods and techniques that provide a clear definition of dental implant systems. In the study, it was planned to develop a system in which domestic brands as well as international brands are included in the system and will recognize the implant brand and system from the patients panoramic radiograph in case of any complications. 12000 implant images with brand, system and diameter information were included in the project. 80% of the images were included in the learning part and 20% were included in the testing part randomly, so that they were balanced across all brands. At the prototype stage, 6 implant systems from 6 different manufacturers were manually marked by the experts with the labeling interface on all panoramic radiographs. The ready data set will be separated as training and test sets and the training set will be trained with CNN architectures. Whereas dentistry was entirely dependent on imported implants in the first years of implant treatments in Turkey, from day to day increasing number of domestic implant production companies are participating in the market. With the increase of both Turkish and foreign implant brands, a need for a successful system that will enable the recognition of the implant system has arisen. Unlike existing systems, the planned algorithm is aimed to recognize the brand of implant systems, as well as their diameters and systems. Thus, a network that correctly classifies the brand, system and diameter of the implant can automatically define the implant system and provide objective information directly to the physician about which components need to be prepared for repair and maintenance when mechanical complications occur. This project is supported by TÜBİTAK with the project number 121E068.

Ikbal Leblebicioglu Kurtulus
Erciyes Üniversitesi Diş Hekimliği Fakültesi
Turkey

Kerem Kılıç
Erciyes Üniversitesi
Turkey

Derviş Karaboğa
Erciyes Üniversitesi Mühendislik Fakültesi
Turkey

Bahriye Akay
Erciyes Üniversitesi Mühendislik Fakültesi

Özkan Ufuk Nalbantoğlu

Alper Baştürk
Erciyes Üniversitesi Mühendislik Fakültesi

Özden Melis Durmaz Yılmaz
Protetik Diş Tedavisi ABD

Serkan Yılmaz

 


Powered by OpenConf®
Copyright ©2002-2021 Zakon Group LLC