All-on-4®全口速定植牙治療方案近年來在臨床端相當盛行，療程時間短、成功率高以及能降低補骨手術的風險。然而，如何快速提供最佳植體擺放位置是個重要的課題，臨床醫師通常依賴經驗評估植體擺放位置，結合有限元素法的最佳化方法所需時間極長，結合深度學習的最佳化方法只考慮單一受力情形，並無考量到其他受力情況。本研究旨在提供一個能取代有限元素法的深度學習模型，能考量不同受力情形並快速提供最佳化方法所需的應力場。
本研究分為三個部分，第一部分為收集數據，建立植體支架有限元素模型並收集不同受力情況下的應力場，第二部分為處理資料，資料型態將從原本的點雲圖轉換為體素圖，並在此過程中選定施力點。第三部為訓練深度學習模型，由於許多因素會影響模型精準度，例如訓練資料量和植體支架種類數量，依照不同的訓練資料量和植體支架種類數量製作不同訓練資料集，將這些訓練資料集拿來訓練模型並找出最佳的訓練資料量和植體支架種類數量，並探討兩者如何影響模型表現。
結果顯示模型最低的平均相對誤差函數值為10.25%，所使用的訓練資料集為10種植體支架模型和3500筆訓練資料。在一般情況下植體支架種類數量和訓練資料量越多，模型的表現會越好，然而如果植體支架種類數量過多會降低每個植體支架所提供的資料量，反而會使模型表現下降，每個植體支架提供300筆資料以上為佳。
本研究開發的深度學習模型能讀取不同的植體支架模型，並預測其在不同受力條件下的應力場，計算時間大約為0.3秒，優於有限元素法所需時間，有望能與最佳化方法結合並大幅縮短最佳化所需時間。

The "All-on-4®" treatment approach, which involves supporting a fixed prosthesis with four dental implants, has become increasingly popular in clinical practice in recent years. However, during the treatment planning process, clinicians often rely on their experience to determine the implant positions without biomechanical considerations. In this study, we propose a Convolutional Neural Network (CNN) model based on 3D U-Net architecture to replace finite element method for predicting biomechanical. In addition, the computational speed of the CNN model is much higher than that of the finite element method, leading to a significant reduction in processing time.
The CNN model comprises convolutional layers, normalization layers, pooling layers, and upsampling layers. It can predict stress field based on different object geometries, force application points, and boundary conditions. To evaluate the performance of CNN model, we compared its results with those obtained using traditional finite element method and used relative error function as our evaluation method. The average relative error of CNN model prediction was found to be 10.25% for stress field and 7.10% for maximum stress value. In particular, the computational time for predicting stress field was significantly reduced from 2 min to 0.3 seconds.
This study presents a novel deep learning model that can replace finite element method for predicting stress field while significantly reducing computational time. In the future, it has the potential to be integrated with optimization to provide clinicians with optimal implant positions.

1. 林家葳, 基於有限元素法的機器學習技術實現All-on-4®贋復設計之即時最佳化, in 機械工程學系. 2022, 國立成功大學: 台南市. p. 99.
2. Maló, P., B. Rangert, and M. Nobre, All-on-4 immediate-function concept with Branemark System® implants for completely edentulous maxillae:: A 1-year retrospective clinical study. Clinical Implant Dentistry and Related Research, 2005. 7: p. S88-S94.
3. Maló, P., et al., "All-on-4" Immediate-Function Concept for Completely Edentulous Maxillae: A Clinical Report on the Medium (3 Years) and Long-Term (5 Years) Outcomes. Clinical Implant Dentistry and Related Research, 2012. 14: p. e139-e150.
4. Malo, P., et al., A longitudinal study of the survival of All-on-4 implants in the mandible with up to 10 Years of follow-up. Journal of the American Dental Association, 2011. 142(3): p. 310-320.
5. Lopes, A., et al., The NobelGuide® All-on-4® Treatment Concept for Rehabilitation of Edentulous Jaws: A Prospective Report on Medium- and Long-Term Outcomes. Clinical Implant Dentistry and Related Research, 2015. 17: p. E406-E416.
6. Wen, W.P., C.Y. Zhang, and C.H. Zhai, Rapid seismic response prediction of RC frames based on deep learning and limited building information. Engineering Structures, 2022. 267.
7. Grigorescu, S., et al., Cloud2Edge Elastic AI Framework for Prototyping and Deployment of AI Inference Engines in Autonomous Vehicles. Sensors, 2020. 20(19).
8. Essien, A. and C. Giannetti, A Deep Learning Model for Smart Manufacturing Using Convolutional LSTM Neural Network Autoencoders. Ieee Transactions on Industrial Informatics, 2020. 16(9): p. 6069-6078.
9. Kim, S., et al., Deep learning-based inverse design for engineering systems: multidisciplinary design optimization of automotive brakes. Structural and Multidisciplinary Optimization, 2022. 65(11).
10. Bhaduri, A., A. Gupta, and L. Graham-Brady, Stress field prediction in fiber-reinforced composite materials using a deep learning approach. Composites Part B-Engineering, 2022. 238.
11. Nastos, C., P. Komninos, and D. Zarouchas, Non-destructive strength prediction of composite laminates utilizing deep learning and the stochastic finite element methods. Composite Structures, 2023. 311.
12. Rao, C.P. and Y. Liu, Three-dimensional convolutional neural network (3D-CNN) for heterogeneous material homogenization. Computational Materials Science, 2020. 184.
13. Liang, L., et al., A deep learning approach to estimate stress distribution: a fast and accurate surrogate of finite-element analysis. Journal of the Royal Society Interface, 2018. 15(138).
14. Bolandi, H., et al., Bridging finite element and deep learning: High-resolution stress distribution prediction in structural components. Frontiers of Structural and Civil Engineering, 2022. 16(11): p. 1365-1377.
15. Dong, G., et al., A part-scale, feature-based surrogate model for residual stresses in the laser powder bed fusion process. Journal of Materials Processing Technology, 2022. 304.
16. Mikula, J., et al., Modelling the influence of process parameters on precipitate formation in powder-bed fusion additive manufacturing of IN718. Materials & Design, 2021. 207.
17. Shin, S., et al., Wheel impact test by deep learning: prediction of location and magnitude of maximum stress. Structural and Multidisciplinary Optimization, 2023. 66(1).
18. Ronneberger, O., P. Fischer, and T. Brox. U-net: Convolutional networks for biomedical image segmentation. in Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18. 2015. Springer.
19. 李祈緯, 結合演化演算法與拓樸最佳化於 All-on-4 全口速定植牙之贋復設計, in 機械工程學系. 2021, 國立成功大學: 台南市. p. 83.
20. Bello, S.A., et al., Review: Deep Learning on 3D Point Clouds. Remote Sensing, 2020. 12(11).
21. Xu, H., et al., Multiscale Multilevel Residual Feature Fusion for Real-Time Infrared Small Target Detection. Ieee Transactions on Geoscience and Remote Sensing, 2023. 61.