21. Diagnosis assistance &
Recommendation of treatment plan
Automated processing
of patient data
(X-ray, CBCT, 3D scan, …)
Automated design
of prosthetic appliance
Deep Learning Applications in Dentistry
AI AI
22. Chen, Y.W. et al., Quintessence Int., 51(3), 2020
Deep Learning Applications in Dentistry
24. Deep Learning Applications in Dentistry
An AI-based technology used in the detection
of possible pathologic lesions.
Suspected caries lesions in the bitewing
radiograph are identified in the software.
Image courtesy of VideaHealth
Chen, Y.W. et al., Quintessence Int., 51(3), 2020
26. Evaluation of Transfer Learning
with Deep Convolutional Neural Networks
for Screening Osteoporosis
in Dental Panoramic Radiographs
27. Deep Convolutional Neural Networks Based Analysis of
Cephalometric Radio Radiographs for Differential Diagnosis
of Orthognathic Surgery Indications
28. Transfer Learning via Deep Convolutional Neural Networks for
Implant Fixture System Classification Using Periapical Radiographs
44. Dice Coefficient : 97.6
Surface Distance : 0.033 mm
Process time: ~ 15s
딥러닝 기반 치과용 3D CT 처리 자동화
45. Deep learning algorithm for
CT segmentation
딥러닝 기반 CT 상하악골 모델링 결과
딥러닝 기반 치과용 3D CT 처리 자동화
상하악골/개별치아/신경관 분할 및 모델링 자동화, 3D 특징점 자동 인식
46. • 3D anatomical landmark detection using CBCT
• Registration of CBCT and 3D dental scan model
High inter- and intra-observer variability
Lots of time and efforts
딥러닝 기반 치과용 CT 데이터 처리 자동화
상하악골/개별치아/신경관 분할 및 모델링 자동화, 3D 특징점 자동 인식
47. 딥러닝 기반 치과용 CT 데이터 처리 자동화
상하악골/개별치아/신경관 분할 및 모델링 자동화, 3D 특징점 자동 인식
CBCT Volume
Create Projections
(Coronal / Sagittal)
Deep Neural Network
…
Initial Detection
…
VOI Centered with
Each Initial Landmarks
…
3D Deep Neural Network Final Detection
First
Stage
Second
Stage
51. 딥러닝 기반 3D 구강 스캔 데이터 처리 자동화
▪ 비정형 데이터 딥러닝을 통한 3D 구강 스캔 데이터의 개별 치아 자동 분할
AI
52. 3D 구강 스캔 & CT 자동 정합 (fully-automated)
딥러닝 기반 3D 구강 스캔, CT 데이터 정합 자동화
53. 딥러닝 기반 3D 구강 스캔, CT 데이터 정합 자동화
Multi-modal Data 정합
54. Multi-Center Dataset for the Registration
▪ External Dataset
▪ CBCT acquired by another institution
▪ Paired CBCT-Full arch scan models
▪ Augmentation for Partially Scanned Dental Model
: 3,304 jaws
▪ Generated by cropping full arch model
56. Registration Accuracy (External Validation)
▪ Average Surface Distance: 0.23 ± 0.14 mm
▪ Average Landmark Distance: 0.42 ± 0.20 mm
▪ Average Processing Time: 4.16 ± 0.55 s
57.
58. 3Dme Studio
Features
- Multi-modality data alignment
- Alignment inspection
- CT data processing
- Coordinate control (to exocad & 3Shape)
- Boolean operation
Bellus3D 3D Facial Scanner
61. 딥러닝 기반 치아 모델 자동생성
3Dme Crown: 교합을 고려한 3D 치아 모델링 자동화 (GAN)
AI
62. 딥러닝 기반 치아 모델 자동생성
3Dme Crown: 교합을 고려한 3D 치아 모델링 자동화 (GAN)
63. 딥러닝 기반 치아 모델 자동생성
3Dme Crown: 교합을 고려한 3D 치아 모델링 자동화 (GAN)
64. 3Dme Crown
▪ AI & Cloud-based crown CAD
▪ Features
- Cloud-based project
- Automated (AI / Library) crown positioning
- Automated AI crown modeling
- Robust mesh handling
65. Internal validation: Training set / Validation set / Test set
Extra-validation: Multi-center Study
RCT (Randomized Controlled Trial) with RWD (Real World Data)
Possibility of Improvement: Transfer Learning, Continuous Learning
AI is only a tool to support human experts!
Efficacy in terms of Time & Cost
AI S/W Performance for Real World Data