近年來牙齒顏色的評估及其再現，仍是牙科美學當中最艱鉅的任務。牙醫師和牙技師須提供外觀自然的贋復物，以滿足患者的美學要求；然而，由於自然牙齒的複雜光學特性，要實現人工製作之贋復物與其顏色和諧仍非易事。使用比色板是臨床上最常用的顏色比對方法。但是，這種方法具有很高的主觀性。環境光源條件、比色者的經驗等因素可能會導致結果不一致。為了解決這個問題，臨床醫師開始採用相關數位比色技術，包括數位攝影、分光光度計、比色計，這些方式客觀、簡單，不但可以減少色差且容易獲得。類神經網路（Artificial neural network）是預測非線性問題的有力工具。其數學模型由一種類似於大腦神經活動的神經元互相連結進行計算。其中，反向傳播神經網路（Back-Propagation Neural Network, BPNN）是一種典型的類神經網絡，已廣泛用於許多醫學領域，例如醫學圖像分析，用於臨床診斷和治療的專門系統，醫學訊號分析和處理。本研究採用了交叉偏振攝影方法拍攝牙齒照片，並使用反向傳播神經網路來預測牙齒顏色，並且將網路預測結果與人眼視覺比色進行比較，為數位牙科的比色提供進一步的建議，以降低牙科在比色方面的不確定性。另外也將神經網路模型運用在氧化鋯試片上，為單層氧化鋯冠的製作上提供建議。
本研究首先將Vita 3D比色板的19個不同顏色色板，透過數位相機對色板進行拍攝，取得色板的CIE L*a*b*值資料集。使用Python Keras框架建構神經網路，由3個輸入變量、2層隱藏層和19個輸出變量組成，其中隱藏層神經元的數量、激勵函數、優化器和批次大小都通過參數測試進行確定。接著透過拍攝模擬牙齒製作而成的試片，將預測結果和人眼比色結果分別與試片計算ΔE值來比較，以驗證此網路演算法的分類效能。此外，也使用同樣的方法建立用來預測氧化鋯試片顏色的網路模型。
結果顯示，將對五個試片比色的結果計算平均值，經過訓練的BPNN模型平均ΔE為1.88±0.22，在臨床可接受的閾值2.65內。然而，視覺比色的平均ΔE為2.93(±1.39)，兩者呈現出顯著的差異(p <0.001）。對於氧化鋯試片，透過一般化測試中發現，色差在臨床可接受閾值內的資料為全部資料的88%，代表有很高比例的預測結果可以被接受且察覺不出差異。
本研究的BPNN模型證明了高準確率的預測，可以進一步幫助牙科臨床醫生為患者選擇合適的贋復物顏色。

Color assessment and reproduction has been the most challenging tasks in esthetic dentistry. However, due to the complex optical characteristics of natural teeth, harmonious color matching of artificial restorations with natural dentition is still difficult to achieve. The present study was aimed at integrating Back-propagation neural network (BPNN) into the process of digital shade matching. Artificial neural network (ANN) models are made of individual processing units called neurons. BPNN is one of the ANN that has been widely used in many medical fields. In this study, The CIE L*a*b* value of 19 different color tabs of the shade guide were taken by a digital camera. The BPNN was constructed by Python and the Keras framework which consisted of three input variables, two hidden layers, and nineteen output variables. Finally, the color difference between the specimens’ color and the model prediction or human visual shade matching were calculated. The result showed that the average ΔE value of BPNN prediction was less than the clinically acceptable threshold. However, the average ΔE value of visual shade matching was higher than that threshold, and there is a significant difference between two methods. The in-house developed BPNN model demonstrated the prediction of high-accuracy and can further help dental clinicians to determine appropriate colors for patients.In this study, The CIE L*a*b* value of 19 different color tabs of the shade guide (Vita 3D-Master) were taken by a digital camera. The BPNN was constructed by Python and the Keras framework which consisted of three input variables, two hidden layers, and nineteen output variables. The parameters of the BPNN, including the number of hidden layer nodes, activation function, optimizer, and batch size, were determined via a parametrical test. Subsequently, specimens simulated to tooth color were passed through the trained BPNN to obtain the predicted output. Finally, the color differences(ΔE) between the specimens’ color and the model prediction or human visual shade matching were calculated. In addition, the same method was used to establish a network model for predicting zirconia specimens.
The result showed that the average ΔE value of BPNN prediction was 1.88 ±0.22, which was less than the clinically acceptable threshold of 2.65(p < 0.001). However, the average ΔE value of visual shade matching was 2.93 ±1.39, and there is a significant difference between two methods. For the zirconia specimens, through the generalization the ΔE value of 88% of the total data was less than the AT value, which means that a high percentage of the prediction results can be accepted, and the difference is not perceive.
In conclusion, the in-house developed BPNN model demonstrated the prediction of high-accuracy and can further help dental clinicians to determine appropriate colors for patients.

1. Bazos, P. and P. Magne, Bio-Emulation: biomimetically emulating nature utilizing a histoanatomic approach; visual synthesis. Int J Esthet Dent, 2014. 9(3): p. 330-52.
2. Sadowsky, S.J., An overview of treatment considerations for esthetic restorations: a review of the literature. J Prosthet Dent, 2006. 96(6): p. 433-42.
3. Seghi, R.R., E.R. Hewlett, and J. Kim, Visual and instrumental colorimetric assessments of small color differences on translucent dental porcelain. J Dent Res, 1989. 68(12): p. 1760-4.
4. Della Bona, A., et al., Visual and instrumental agreement in dental shade selection: three distinct observer populations and shade matching protocols. Dent Mater, 2009. 25(2): p. 276-81.
5. Sarafianou, A., et al., Matching repeatability and interdevice agreement of 2 intraoral spectrophotometers. The Journal of prosthetic dentistry, 2012. 107(3): p. 178-185.
6. Jarad, F.D., M.D. Russell, and B.W. Moss, The use of digital imaging for colour matching and communication in restorative dentistry. Br Dent J, 2005. 199(1): p. 43-9; discussion 33.
7. Fried, D., et al., Nature of light scattering in dental enamel and dentin at visible and near-infrared wavelengths. Applied optics, 1995. 34(7): p. 1278-1285.
8. Ten Bosch, J. and J. Coops, Tooth color and reflectance as related to light scattering and enamel hardness. Journal of dental research, 1995. 74(1): p. 374-380.
9. He, W.H., et al., Evaluating the relationship between tooth color and enamel thickness, using twin flash photography, cross‐polarization photography, and spectrophotometer. Journal of Esthetic and Restorative Dentistry, 2020. 32(1): p. 91-101.
10. Nakhaei, M., et al., The influence of dental shade guides and experience on the accuracy of shade matching. The journal of contemporary dental practice, 2016. 17(1): p. 22-26.
11. Chitrarsu, V.K., A.S. Chidambaranathan, and M. Balasubramaniam, Analysis of shade matching in natural dentitions using intraoral digital spectrophotometer in LED and filtered LED light sources. Journal of Prosthodontics, 2019. 28(1): p. e68-e73.
12. Ragain, J.C. A Review of Color Science in Dentistry: Shade Matching in the Contemporary Dental Practice. 2016.
13. Curd, F.M., et al., Comparison of the shade matching ability of dental students using two light sources. The Journal of prosthetic dentistry, 2006. 96(6): p. 391-396.
14. Corcodel, N., et al., Effect of external light conditions during matching of tooth color: an intraindividual comparison. International Journal of Prosthodontics, 2009. 22(1).
15. Jasinevicius, T.R., et al., Shade‐matching abilities of dental laboratory technicians using a commercial light source. Journal of Prosthodontics: Implant, Esthetic and Reconstructive Dentistry, 2009. 18(1): p. 60-63.
16. Wee, A.G., et al., Lighting conditions used during visual shade matching in private dental offices. The Journal of prosthetic dentistry, 2016. 115(4): p. 469-474.
17. Lee, Y.-K., et al., Perceived color shift of a shade guide according to the change of illuminant. The Journal of Prosthetic Dentistry, 2011. 105(2): p. 91-99.
18. Gokce, H.S., et al., Shade matching performance of normal and color vision-deficient dental professionals with standard daylight and tungsten illuminants. J Prosthet Dent, 2010. 103(3): p. 139-47.
19. Çapa, N., E. Kazazoğlu, and S. Çalikkocaoğlu, Evaluating factors that affect the shade-matching ability of dentists, dental staff members and laypeople. The Journal of the American Dental Association, 2010. 141(1): p. 71-76.
20. Horn, D.J., J. Bulan-Brady, and M.L. Hicks, Sphere spectrophotometer versus human evaluation of tooth shade. Journal of endodontics, 1998. 24(12): p. 786-790.
21. Ragain, J.C., A review of color science in dentistry: colorimetry and color space. J Dent Oral Disord Ther, 2016. 4(1): p. 1-5.
22. O'Brien, W.J., K.M. Boenke, and C.L. Groh, Coverage errors of two shade guides. International Journal of Prosthodontics, 1991. 4(1).
23. Rubino, M., et al., Colour measurement of human teeth and evaluation of a colour guide. Color Research & Application, 1994. 19(1): p. 19-22.
24. Mahn, E., et al., Comparison of visual analog shade matching, a digital visual method with a cross-polarized light filter, and a spectrophotometer for dental color matching. The Journal of prosthetic dentistry, 2021. 125(3): p. 511-516.
25. MINOLTA, K., Precise color Communication: Color control from perception to instrumentation. 1998. 2015.
26. Paravina, R.D. and J.M. Powers, Esthetic Color Training in Dentistry (Interactive CD-ROM Inside). 2004.
27. Kim-Pusateri, S., et al., Reliability and accuracy of four dental shade-matching devices. The Journal of prosthetic dentistry, 2009. 101(3): p. 193-199.
28. Paul, S.J., et al., Conventional visual vs spectrophotometric shade taking for porcelain-fused-to-metal crowns: a clinical comparison. Journal of Prosthetic Dentistry, 2004. 92(6): p. 577.
29. Khurana, R., et al., A clinical evaluation of the individual repeatability of three commercially available colour measuring devices. British dental journal, 2007. 203(12): p. 675-680.
30. Paul, S., et al., Visual and spectrophotometric shade analysis of human teeth. Journal of dental research, 2002. 81(8): p. 578-582.
31. Dozić, A., et al., Performance of five commercially available tooth color‐measuring devices. Journal of Prosthodontics, 2007. 16(2): p. 93-100.
32. Jivanescu, A., et al., Conventional versus spectrophotometric shade taking for the upper central incisor: a clinical comparative study. Timisoara Medical Journal, 2010. 60: p. 274-279.
33. Jarad, F., M. Russell, and B. Moss, The use of digital imaging for colour matching and communication in restorative dentistry. British dental journal, 2005. 199(1): p. 43-49.
34. Wee, A.G., et al., Color accuracy of commercial digital cameras for use in dentistry. Dental Materials, 2006. 22(6): p. 553-559.
35. Tam, W. and H. Lee, Dental shade matching using a digital camera. Journal of dentistry, 2012. 40: p. e3-e10.
36. Sajjadi, S.H., et al., Effects of Three Types of Digital Camera Sensors on Dental Specialists’ Perception of Smile Esthetics: A Preliminary Double‐Blind Clinical Trial. Journal of Prosthodontics, 2016. 25(8): p. 675-681.
37. Casaglia, A., et al., Dental photography today. Part 1: basic concepts. ORAL & implantology, 2015. 8(4): p. 122.
38. Sampaio, C.S., et al., Variability of color matching with different digital photography techniques and a gray reference card. The Journal of prosthetic dentistry, 2019. 121(2): p. 333-339.
39. Hein, S., J. Tapia, and P. Bazos, eLABor_aid: a new approach to digital shade management. Int J Esthet Dent, 2017. 12(2): p. 186-202.
40. Clary, J.A., et al., Influence of light source, polarization, education, and training on shade matching quality. The Journal of prosthetic dentistry, 2016. 116(1): p. 91-97.
41. Gurrea, J., et al., Evaluation of Dental Shade Guide Variability Using Cross-Polarized Photography. International Journal of Periodontics & Restorative Dentistry, 2016. 36(5).
42. Mine, Y., et al., Applying deep artificial neural network approach to maxillofacial prostheses coloration. Journal of prosthodontic research, 2020. 64(3): p. 296-300.
43. Lee, J.-H., et al., Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. Journal of dentistry, 2018. 77: p. 106-111.
44. Lee, J.-H., et al., Diagnosis and prediction of periodontally compromised teeth using a deep learning-based convolutional neural network algorithm. Journal of periodontal & implant science, 2018. 48(2): p. 114.
45. Wang, C.-W., et al., Evaluation and comparison of anatomical landmark detection methods for cephalometric x-ray images: a grand challenge. IEEE transactions on medical imaging, 2015. 34(9): p. 1890-1900.
46. Sawagashira, T., et al., An automatic detection method for carotid artery calcifications using top-hat filter on dental panoramic radiographs. IEICE TRANSACTIONS on Information and Systems, 2013. 96(8): p. 1878-1881.
47. Miki, Y., et al., Classification of teeth in cone-beam CT using deep convolutional neural network. Computers in biology and medicine, 2017. 80: p. 24-29.
48. Xie, X., L. Wang, and A. Wang, Artificial neural network modeling for deciding if extractions are necessary prior to orthodontic treatment. The Angle orthodontist, 2010. 80(2): p. 262-266.
49. Kim, D.W., et al., Machine learning to predict the occurrence of bisphosphonate-related osteonecrosis of the jaw associated with dental extraction: A preliminary report. Bone, 2018. 116: p. 207-214.
50. Jayalakshmi, T. and A. Santhakumaran, Statistical normalization and back propagation for classification. International Journal of Computer Theory and Engineering, 2011. 3(1): p. 1793-8201.
51. Sedda, M., et al., Influence of coloring procedure on flexural resistance of zirconia blocks. The Journal of prosthetic dentistry, 2015. 114(1): p. 98-102.
52. Davidowitz, G. and P.G. Kotick, The use of CAD/CAM in dentistry. Dental Clinics of North America, 2011. 55(3): p. 559-70, ix.
53. Yilmaz, H., C. Aydin, and B.E. Gul, Flexural strength and fracture toughness of dental core ceramics. The Journal of prosthetic dentistry, 2007. 98(2): p. 120-128.
54. Seghi, R.R. and J.A. Sorensen, Relative flexural strength of six new ceramic materials. International Journal of Prosthodontics, 1995. 8: p. 239-239.
55. Teixeira, E.C., et al., Dynamic fatigue and strength characterization of three ceramic materials. Journal of Materials Science: Materials in Medicine, 2007. 18(6): p. 1219-1224.
56. Tinschert, J., et al., Fracture Resistance of Lithium Disilicate--, Alumina-, and Zirconia-Based Three-Unit Fixed Partial Dentures: A Laboratory Study. International Journal of Prosthodontics, 2001. 14(3).
57. Heffernan, M.J., et al., Relative translucency of six all-ceramic systems. Part II: core and veneer materials. The Journal of prosthetic dentistry, 2002. 88(1): p. 10-15.
58. Ilie, N. and B. Stawarczyk, Quantification of the amount of light passing through zirconia: the effect of material shade, thickness, and curing conditions. Journal of dentistry, 2014. 42(6): p. 684-690.
59. Zhang, Y., Making yttria-stabilized tetragonal zirconia translucent. Dental materials, 2014. 30(10): p. 1195-1203.
60. Alaniz, J., et al., Optical properties of transparent nanocrystalline yttria stabilized zirconia. Optical Materials, 2009. 32(1): p. 62-68.
61. Rinke, S. and C. Fischer, Range of indications for translucent zirconia modifications: clinical and technical aspects. Quintessence International, 2013. 44(8).
62. Miyazaki, T., et al., Current status of zirconia restoration. Journal of prosthodontic research, 2013. 57(4): p. 236-261.
63. Li, D.-c., et al., Effect of cerium oxide and erbium oxide as colorants on the chromaticity and mechanical properties of dental zirconia ceramic. Journal of Clinical Rehabilitative Tissue Engineering Research, 2011. 15(47): p. 8829-8832.
64. Shah, K., J. Holloway, and I. Denry, Effect of coloring with various metal oxides on the microstructure, color, and flexural strength of 3Y‐TZP. Journal of Biomedical Materials Research Part B: Applied Biomaterials: An Official Journal of The Society for Biomaterials, The Japanese Society for Biomaterials, and The Australian Society for Biomaterials and the Korean Society for Biomaterials, 2008. 87(2): p. 329-337.
65. Hjerppe, J., et al., Effect of shading the zirconia framework on biaxial strength and surface microhardness. Acta Odontologica Scandinavica, 2008. 66(5): p. 262-267.
66. Aboushelib, M.N., et al., The influence of pigments on the slow crack growth in dental zirconia. Dental Materials, 2012. 28(4): p. 410-415.
67. Tabatabaian, F., Color aspect of monolithic zirconia restorations: a review of the literature. Journal of prosthodontics, 2019. 28(3): p. 276-287.
68. Ragain, J.C., Matching the optical properties of direct esthetic dental restorative materials to those of human enamel and dentin. 1998, The Ohio State University.
69. Ragain, J.C., A review of color science in dentistry: The process of color vision. J Dent Oral Disord Ther, 2015. 3(01): p. 1-4.
70. CId, E.c., CIE Technical Report: Colorimetry. CIE Pub No. 15.3. 2004, CIE Central Bureau Vienna, Austria.
71. Pecho, O.E., et al., Visual and instrumental shade matching using CIELAB and CIEDE2000 color difference formulas. Dental materials, 2016. 32(1): p. 82-92.
72. Paravina, R.D., et al., Color difference thresholds in dentistry. Journal of Esthetic and Restorative Dentistry, 2015. 27: p. S1-S9.
73. Van Rossum, G. and F.L. Drake Jr, Python tutorial. Vol. 620. 1995: Centrum voor Wiskunde en Informatica Amsterdam.
74. Abadi, M., et al. Tensorflow: A system for large-scale machine learning. in 12th {USENIX} symposium on operating systems design and implementation ({OSDI} 16). 2016.
75. Zhang, Z. Improved adam optimizer for deep neural networks. in 2018 IEEE/ACM 26th International Symposium on Quality of Service (IWQoS). 2018. IEEE.
76. Sadowski, P., Notes on backpropagation. homepage: https://www. ics. uci. edu/pjsadows/notes. pdf (online), 2016.
77. Anguita, D., et al. The'K'in K-fold Cross Validation. in ESANN. 2012.
78. Bengio, Y. and Y. Grandvalet, No unbiased estimator of the variance of k-fold cross-validation. Journal of machine learning research, 2004. 5(Sep): p. 1089-1105.
79. Wei, J., et al., Evaluation of a Novel Computer Color Matching System Based on the Improved Back‐Propagation Neural Network Model. Journal of Prosthodontics, 2018. 27(8): p. 775-783.
80. Brownlee, J., What is the Difference Between a Batch and an Epoch in a Neural Network? Deep Learning; Machine Learning Mastery: Vermont, VIC, Australia, 2018.
81. Keskar, N.S., et al., On large-batch training for deep learning: Generalization gap and sharp minima. arXiv preprint arXiv:1609.04836, 2016.
82. Goyal, P., et al., Accurate, large minibatch sgd: Training imagenet in 1 hour. arXiv preprint arXiv:1706.02677, 2017.
83. Hoffer, E., I. Hubara, and D. Soudry, Train longer, generalize better: closing the generalization gap in large batch training of neural networks. arXiv preprint arXiv:1705.08741, 2017.
84. Reed, R. and R.J. MarksII, Neural smithing: supervised learning in feedforward artificial neural networks. 1999: Mit Press.
85. Goodfellow, I., et al., Deep learning. Vol. 1. 2016: MIT press Cambridge.
86. Hu, S., et al., A review about building hidden layer methods of deep learning. Journal of Advances in Information Technology Vol, 2016. 7(1).
87. Stathakis, D., How many hidden layers and nodes? International Journal of Remote Sensing, 2009. 30(8): p. 2133-2147.
88. Agarap, A.F., Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375, 2018.
89. Gómez-Polo, C., et al., Comparison of the CIELab and CIEDE2000 color difference formulas. The journal of prosthetic dentistry, 2016. 115(1): p. 65-70.
90. Li, Q. and Y. Wang, Comparison of shade matching by visual observation and an intraoral dental colorimeter. Journal of oral rehabilitation, 2007. 34(11): p. 848-854.
91. Schropp, L., Shade matching assisted by digital photography and computer software. Journal of Prosthodontics: Implant, Esthetic and Reconstructive Dentistry, 2009. 18(3): p. 235-241.
92. Tam, W.-K. and H.-J. Lee, Accurate shade image matching by using a smartphone camera. Journal of prosthodontic research, 2017. 61(2): p. 168-176.
93. Kim, M., et al., A digital shade-matching device for dental color determination using the support vector machine algorithm. Sensors, 2018. 18(9): p. 3051.
94. Justiawan, D.A.W., et al., Comparative Analysis of Color Matching System for Teeth Recognition Using Color Moment. Medical Devices (Auckland, NZ), 2019. 12: p. 497.