Computer vision and image processing for dental products
3 likes1,262 views
The document discusses the application of data mining in aesthetic dentistry through a platform called Visagismile, which seeks to automate facial type classification for improved dental design. It outlines the challenges of transitioning from 2D projections to accurate 3D models for dental aesthetics and presents machine learning methods to enhance classification accuracy. Additionally, it introduces Rebel.dental, a forthcoming product aimed at streamlining digital smile design processes.
Computer vision and image processing for dental products
- 1. Discussion on Computer Vision and Image Processing: Applying data mining in aesthetic dentistry Boyan Yankov, PhD
- 2. About me Web developer Co-founder of Web Motion Co-founder of VisagiSMile Actively working on innovative dental software during the last 2 years PhD in Informatics, Technology Entrepreneurship
- 3. Introduction
- 4. VisagiSMile - our first dental product 2000+ dentists 4000+ clinical cases
- 5. The Visagism Concept - Facial Perception and Teeth Structure Do you believe these designs represent the following effects? YES / NO / MAYBE
- 6. The Visagism Concept - Facial Type Example: Patients who have rectangular faces, formed by well-defined angles and vertical and horizontal lines, have a visual expression of strength and power, are generally related to the strong facial type. Dynamic / Strong
- 7. The Visagism Concept - Facial Type Facial shapes Strong Dynamic Delicate Calm The facial type depends on the facial shapes, which are defined with a facial map. Delicate / Dynamic
- 8. The Visagism Concept - Personality Hippocrates, 460-370 BC Dr. Susan Dellinger’s test and Eysenk’s personality questionnaires StrongDynamic Calm Delicate
- 9. The Visagism Concept The dental visagism theory relates: Facial type and personality Teeth structure
- 12. Problem: Classifying the Facial Type A skilled dental technician can subjectively determine the facial type. How to automate this process with a high accuracy?
- 13. Classifying the Facial Type - Data Collection A dental visagism professional has classified the types of 91 faces manually. And has selected the corresponding facial shapes.
- 14. Classifying the Facial Type - Data Collection Software tool for manual facial shape selection
- 15. Classifying the Facial Type - Data Collection Automated facial shape selection 1. Facial image is uploaded by the user 2. API call to Betaface is made (with JavaScript) 3. Betaface identifies the Face - provides the positions of the landmark points 4. The face is calibrated by using the bipupilar line (with JavaScript) 5. The facial map is calculated (with Javascript) 6. ...
- 16. Classifying the Facial Type - by Geometrical Approximation The shapes of different facial types look the same until overlapped.
- 17. Classifying the Facial Type - by Geometrical Approximation Which shape does the facial contour fit best (geometrically)? Disadvantages: Do all points matter? Are all points equally important? 56% successful classifications (initial algorithm)
- 18. Classifying the Facial Type - by Applying Machine Learning
- 19. Classifying the Facial Type - IBM SPSS Modeler Testing Setup
- 20. Classifying the Facial Type - IBM SPSS Modeler Results
- 21. Classifying the Facial Type - IBM SPSS Modeler Results C5.1 Classification tree model - top fragment:
- 22. Classifying the Facial Type - Weka Results === Run information === Scheme:weka.classifiers.functions.Logistic -R 1.0E-8 -M -1 Correctly Classified Instances 46 49.4624 % Incorrectly Classified Instances 47 50.5376 % Kappa statistic 0.3066 Mean absolute error 0.2874 Root mean squared error 0.4376 Relative absolute error 78.6179 % Root relative squared error 102.3622 % Total Number of Instances 93
- 23. Product Validation and Clinical Cases BEFORE PORCELAIN VENEERS DESIGN AFTER
- 24. Problem: From 2D projection to 3D models FROM THIS...
- 25. Problem: From 2D projection to 3D models ...TO THIS
- 26. From 2D projection to 3D models ● 60 cases ● All pictures and measurements taken according to clinical protocol ● Analysis of upper teeth: ○ Central incisors (11, 21) ○ Lateral incisors (12, 22) ○ Canines (13, 23) ● Goal: Width in projection (photo) to width of 3D scan ● Goal: Height in projection (photo) to height of 3D scan
- 27. From 2D projection to 3D models
- 28. From 2D projection to 3D models
- 29. From 2D projection to 3D models
- 30. From 2D projection to 3D models
- 31. From 2D projection to 3D models - conclusions Central incisors ● Height can be determined from the model. (R Square = 0,648) ● Width can be determined from the model, but less accurately (R Square = 0,464) Lateral incisors ● Height can be determined from the model. (R Square = 0,752) ● Width can be determined from the model, but less accurately (R Square = 0,294) Canines ● Height can be determined from the model. (R Square = 0,761) ● Width can not be accurately determined from the model (R Square = 0,054) Best R square values for all teeth types have been calculated for the quadratic model, y = a + b1*x + b2*x^2
- 32. From 2D projection to 3D models - Software ● Chooses tooth shape according to the tooth type and to the personalized smile design ● Calculates the width and height of the tooth in 3D from the width and height in projection ● Resizes the 3D model by using a custom 3D library ● Visualizes the tooth with controls for zoom and rotation ● Exports the 3D model in STL format
- 33. REBEL.DENTAL - Coming soon 1. Provides seamless and inexpensive digital smile design 2. Produces a realistic 3D teeth mock-up - ready for 3D printing 3. Works with local digital labs 4. Fully automates the digital smile design process
- 34. REBEL.DENTAL
- 35. Product Validation and Clinical Cases - Design
- 36. Product Validation and Clinical Cases - Intraoral scan
- 37. Product Validation and Clinical Cases - Mock up
- 38. Product Validation and Clinical Cases BEFORE TREATMENT AFTER TREATMENT
- 39. Product Validation and Clinical Cases BEFORE TREATMENT AFTER TREATMENT
- 40. Product Validation and Clinical Cases - Another Case AFTER TREATMENT
- 41. Free trial Register a account at visagismile.com Rebel.dental is coming soon!
- 42. Thank you for your attention! I will be happy to answer any questions you may have. Contacts: Boyan Yankov boyan@webmotion.bg www.webmotion.bg