Earticle

초록

영어

Purpose: This study aimed to generate virtual mandibular first molars using deep learning technology, specifically deep convolutional generative adversarial network (DCGAN), and evaluate the accuracy and reliability of these virtual teeth by analyzing their morphological characteristics. These morphological characteristics were classified based on various evaluation criteria, facilitating the assessment of deep learning-based dental prosthesis production’s practical applicability. Methods: Based on our previous research, 1,000 virtual mandibular first molars were generated, and based on morphological criteria, categorized as matching, non-matching, and partially matching. The generated first molars or the categorization of the generated molars were analyzed through the expert judgment of dental technicians. Results: Among the 1,000 generated virtual teeth, 143 (14.3%) met all five evaluation criteria, whereas 76 (7.6%) were judged as completely non-matching. The most frequent issue, with 781 (78.1%) instances, including some overlapping instances, was related to occlusal buccal cusp discrepancies. Conclusion: The study reveals the potential of DCGAN-generated virtual teeth as substitutes for real teeth; however, additional research and improvements in data quality are necessary to enhance accuracy. Continued data collection and refinement of generation methods can maximize the practicality and utility of deep learning-based dental prosthesis production.

목차

Abstract
INTRODUCTION
MATERIALS AND METHODS
1. 데이터 셋 구성
2. 형태학적 분석
RESULTS
DISCUSSION
CONCLUSIONS
FUNDING
ACKNOWLEDGEMENTS
CONFLICT OF INTEREST
ORCID
REFERENCES

키워드

저자

• 배은정 [ Eun-Jeong Bae | 부천대학교 치기공과 ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  대한치과기공학회지 [THE JOURNAL OF KOREAN ACADEMY OF DENTAL TECHNOLOGY]

• 간기

  계간

• pISSN

  1229-3954

• eISSN

  2288-5218

• 수록기간

  1979~2026

• 등재여부

  KCI 등재

• 십진분류

  KDC 515 DDC 617