All-on-4®全口重建治療已廣泛應用於大量缺牙患者，其植體配置方式會直接影響植體周圍骨組織之生物力學表現，因此如何快速且準確地規劃患者個別植體配置已成為重要研究議題。有限元素分析雖可提供可靠之生物力學評估結果，但於植體配置最佳化過程中需反覆建立模型與求解，造成極高之運算成本。此外，僅利用標準顎骨有限元素資料建立之人工智慧模型，亦難以充分反映不同患者間之個體差異，而患者個別有限元素資料取得不易，使建立個人化預測模型面臨少樣本問題。
為解決上述問題，本研究提出一套結合有限元素分析、少樣本微調、多輸出人工神經網路及粒子群最佳化之 All-on-4®植體配置最佳化流程。首先利用標準顎骨有限元素模擬資料建立基礎模型，再以 9 位真實患者之下顎 CBCT 重建模型進行患者個別有限元素分析，利用少量患者資料完成模型微調，以建立患者個別代理模型。最後結合粒子群最佳化演算法快速搜尋患者個別植體配置，並以有限元素分析驗證最佳化結果。
研究結果顯示，相較於直接使用基礎模型進行預測，利用少量患者個別有限元素資料進行模型微調後，可明顯提升患者個別模型之預測能力。相較於基礎模型在9位患者上平均有23.87%的預測誤差，使用10筆資料進行微調的患者個別模型，預測誤差可下降至9.23%。於植體配置最佳化方面，FSMONN-PSO 所得之最佳解與有限元素最佳化所得之最佳解，在 9 位患者上的誤差為 1.68%~15.63%。而模型最佳植體配置與有限元素最佳化的結果相近。
本研究所提出之方法能大幅縮短傳統有限元素分析結合最佳化流程所需之運算時間，可作為未來患者個別 All-on-4®植體配置規劃之基礎架構，並具有發展為臨床治療規劃決策支援工具。

The All-on-4® concept has been widely applied for full-arch rehabilitation in patients with extensive tooth loss. Implant configuration directly affects the biomechanical behavior of peri-implant bone. Although finite element analysis provides reliable biomechanical evaluations, repeated simulations during implant optimization result in substantial computational costs. Furthermore, artificial intelligence models trained solely on standard mandibular data may not adequately account for patient-specific variations, while obtaining sufficient patient-specific FEA data remains challenging.
This study proposes an All-on-4® implant configuration optimization framework integrating FEA, few-shot fine-tuning, a multi-output artificial neural network, and particle swarm optimization (PSO). A multi-output base model was first developed using FEA data from a standard mandible. Patient-specific mandibular models were then reconstructed from cone-beam computed tomography images of nine patients. A small number of patient-specific FEA samples were used to fine-tune the base model and establish patient-specific surrogate models. These models were integrated with PSO to identify optimal implant configurations, which were subsequently validated using FEA.
The results showed that few-shot fine-tuning substantially improved patient-specific prediction accuracy. Across nine patients, the mean prediction error decreased from 23.87% with the base model to 9.23% after fine-tuning with only 10 patient-specific samples. For implant optimization, the optimal solutions obtained using FSMONN-PSO showed errors of 1.68%–15.63% compared with those obtained using FEA-PSO, while producing similar optimal implant configurations.
The proposed framework reduces the computational cost of conventional FEA-based optimization while accounting for patient-specific variations. It provides a potential foundation for patient-specific All-on-4® implant planning and future clinical decision-support tools.

[1]Maló, P., Rangert, B. & Nobre, M. “All-on-Four” Immediate-Function Concept with Brånemark System® Implants for Completely Edentulous Mandibles: A Retrospective Clinical Study. Clinical Implant Dentistry and Related Research, 5(Suppl 1), 2–9 (2003).
[2]Ayali, A. et al. Biomechanical comparison of the All-on-4, M-4, and V-4 techniques in an atrophic maxilla: A 3D finite element analysis. Computers in Biology and Medicine, 123, 10 (2020).
[3]Gumrukcu, Z., Korkmaz, Y. T. & Korkmaz, F. M. Biomechanical evaluation of implant-supported prosthesis with various tilting implant angles and bone types in atrophic maxilla: A finite element study. Computers in Biology and Medicine, 86, 47–54 (2017).
[4]de Sousa, A. A. & Mattos, B. S. C. Finite element analysis of stability and functional stress with implant-supported maxillary obturator prostheses. Journal of Prosthetic Dentistry, 112(6), 1578–1584 (2014).
[5]Moreira de Melo, E. J. M. & Francischone, C. E. Three-dimensional finite element analysis of two angled narrow-diameter implant designs for an all-on-4 prosthesis. Journal of Prosthetic Dentistry, 124(4), 477–484 (2020).
[6]Horita, S. et al. Biomechanical analysis of immediately loaded implants according to the “All-on-Four” concept. Journal of Prosthodontic Research, 61(2), 123–132 (2017).
[7]Dogan, D. O. et al. Evaluation of “All-on-Four” Concept and Alternative Designs with 3D Finite Element Analysis Method. Clinical Implant Dentistry and Related Research, 16(4), 501–510 (2014).
[8]Chen, Y. C. et al. Real-time optimization of prosthetic design for complete arch implant-supported treatments using finite element-based machine learning. Journal of Prosthetic Dentistry,134(5), 1920-1928 (2025)
[9]Maló, P., Rangert, B. & Nobre, M. All-on-4 immediate-function concept with Brånemark System implants for completely edentulous maxillae: A 1-year retrospective clinical study. Clinical Implant Dentistry and Related Research, 7(Suppl 1), S88–S94 (2005).
[10]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, 14(Suppl 1), e139–e150 (2012).
[11]Maló, P. et al. The All-on-4 treatment concept for the rehabilitation of the completely edentulous mandible: A longitudinal study with 10 to 18 years of follow-up. Clinical Implant Dentistry and Related Research, 21(4), 565–577 (2019).
[12]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, 17, E406–E416 (2015).
[13]Bevilacqua, M. et al. Three-dimensional finite element analysis of load transmission using different implant inclinations and cantilever lengths. Int. J. Prosthodont., 21(6), 539–542 (2008).
[14]Gupta, Y. et al. Design of dental implant using design of experiment and topology optimization: A finite element analysis study. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine, 235(2), 157–166 (2020).
[15]Anitua, E. & Orive, G. Finite Element Analysis of the Influence of the Offset Placement of an Implant-Supported Prosthesis on Bone Stress Distribution. Journal of Biomedical Materials Research Part B-Applied Biomaterials, 89B(2), 275–281 (2009).
[16]Skiedraite, I., Diliunas, S. & Varinauskas, V. A Study Case of Short Dental Implants Loading in the All-On-4 System on Fixed Dental Prostheses: A Finite Element Analysis. Mechanika, 29(2), 104–109 (2023).
[17]Ortiz-Puigpelat, O., Lázaro-Abdulkarim, A., de Medrano-Reñé, J. M., Gargallo-Albiol, J., Cabratosa-Termes, J. & Hernández-Alfaro, F. Influence of Implant Position in Implant-Assisted Removable Partial Denture: A Three-Dimensional Finite Element Analysis. Journal of Prosthodontics, 28(2), e675–e681 (2019).
[18]Chanda, S., Gupta, S. & Pratihar, D. K. Effects of interfacial conditions on shape optimization of cementless hip stem: An investigation based on a hybrid framework. Struct. Multidisc. Optim., 53, 1143–1155 (2016).
[19]Roy, S. et al. Design of patient specific dental implant using FE analysis and computational intelligence techniques. Appl. Soft Comput., 65, 272–279 (2018).
[20]Phellan, R. et al. Real-time biomechanics using the finite element method and machine learning: Review and perspective. Med. Phys., 48, 7–18 (2021).
[21]Rodero, C., Baptiste, T. M. G., Barrows, R. K., Lewalle, A., Niederer, S. A. & Strocchi, M. Advancing clinical translation of cardiac biomechanics models: A comprehensive review, applications and future pathways. Front. Phys., 11, 1306210 (2023).
[22]Vecchiato, M. et al. Artificial intelligence applications in sport-related concussion: An updated scoping review. Journal of Science and Medicine in Sport (2026).
[23]Tuladhar, U. R., Simon, R., Mix, D. & Richards, M. S. 2D ultrasound elasticity imaging of abdominal aortic aneurysms using deep neural networks. IEEE Trans. Comput. Imaging, 12, 921–934 (2026).
[24]Shi, L., Chen, Y. & Vedula, V. HeartSimSage: Attention-enhanced graph neural networks for accelerating cardiac mechanics modeling. J. Comput. Phys., 560, 114895 (2026).
[25]Shao, H. et al. Real-time liver tumor localization via a single x-ray projection using deep graph neural network-assisted biomechanical modeling. Phys. Med. Biol., 67(11), 115009 (2022).
[26]Sajjadinia, S. et al. Multi-fidelity surrogate modeling through hybrid machine learning for biomechanical and finite element analysis of soft tissues. Computers in Biology and Medicine, 148, 105699 (2022).
[27]Saillard, E. et al. Finite element models with automatic computed tomography bone segmentation for failure load computation. Scientific Reports, 14, 16576 (2024).
[28]Rodriguez-Molinero, J. & Prados-Privado, M. Time-resolved prediction of dental implant biomechanics through integration of finite element analysis, osseointegration dynamics, and deep learning. J. Mech. Behav. Biomed. Mater., 175, 107316 (2026).
[29]Rengarajan, B., Patnaik, S. S. & Finol, E. A. A predictive analysis of wall stress in abdominal aortic aneurysms using a neural network model. Journal of Biomechanical Engineering, 143(12), 121004 (2021).
[30]Rahman, E. et al. AI-driven insights into glabellar wrinkle patterns: Reassessing the standardised botulinum toxin: An injection protocol to address anatomical variability. Aesthet. Plast. Surg. (2025).
[31]Mohammadi, H. & Arjmand, N. An artificial neural network to estimate detailed active–passive spinal loads during static lifting activities for a standardized anthropometry. J. Biomech., 195, 113115 (2026).
[32]Minku & Ghosh, R. A Macro–Micro FE and Machine Learning Based Design of Diamond Lattice Tibial Implant to Improve Biomechanical and Osseointegration Performance. Int. J. Numer. Methods Biomed. Eng., 41, e70133 (2025).
[33]Maag, C., Fitzpatrick, C. K. & Rullkoetter, P. J. Evaluation of machine learning techniques for real-time prediction of implanted lower limb mechanics. Front. Bioeng. Biotechnol., 12, 1461768 (2025).
[34]Ma, X. et al. A geometric deep learning model for real-time prediction of knee joint biomechanics under meniscal extrusion. Ann. Biomed. Eng., 53, 2503–2512 (2025).
[35]Lin, R. & Zhang, J. A novel approach to biomechanical modeling: CT image weight initialization and Physics Informed Neural Networks. Biomed. Signal Process. Control, 109, 107939 (2025).
[36]Liang, L., Liu, M., Elefteriades, J. & Sun, W. Synergistic Integration of Deep Neural Networks and Finite Element Method with Applications for Biomechanical Analysis of Human Aorta. bioRxiv (2023).
[37]Liew, B. X. W. et al. Comparing shallow, deep, and transfer learning in predicting joint moments in running. J. Biomech., 129, 110820 (2021).
[38]Lampen, N. et al. Deep learning for biomechanical modeling of facial tissue deformation in orthognathic surgical planning. Int. J. Comput. Assist. Radiol. Surg., 17(5), 945–952 (2022).
[39]Jiang, J. et al. Real-time simulation for multi-component biomechanical analysis using localized tissue constraint progressive transfer learning. J. Mech. Behav. Biomed. Mater., 158, 106682 (2024).
[40]Haribabu, G. N. & Basu, B. Implementing machine learning approaches for accelerated prediction of bone strain in acetabulum of a hip joint. J. Mech. Behav. Biomed. Mater., 153, 106495 (2024).
[41]Goswami, S. et al. Neural operator learning of heterogeneous mechanobiological insults contributing to aortic aneurysms. J. R. Soc. Interface, 19(194), 20220410 (2022).
[42]Zou, J. et al. Prediction on the medial knee contact force in patients with knee valgus using transfer learning approaches: Application to rehabilitation gaits. Comput. Biol. Med., 150, 106099 (2022).
[43]Chung, T. K., Liang, N. L. & Vorp, D. A. Artificial intelligence framework to predict wall stress in abdominal aortic aneurysm. Appl. Eng. Sci., 10, 100104 (2022).
[44]Chen, Y. et al. Deep learning-based estimation of myocardial material parameters from cardiac MRI. Bioengineering, 12(4), 433 (2025).
[45]Kearney, K. M. et al. From simulation to reality: Predicting torque with fatigue onset via transfer learning. IEEE Trans. Neural Syst. Rehabil. Eng., 32, 3669–3676 (2024).
[46]Chen, X., El-Bouri, W., Payne, S. & Lu, L. Prediction of post-stroke brain swelling using biomechanical modelling and deep neural networks. Med. Image Anal., 111, 104067 (2026).
[47]Azadi, A. et al. Deep learning model for maximum principal strain prediction from ice hockey video-derived impact features. Sports Eng., 29, 2 (2026).
[48]Zhang, L., Cao, Z. & Zhao, Q. Deep learning-aided segmentation combined with finite element analysis reveals a more natural biomechanic of dinosaur fossil. Sci. Rep., 15, 13964 (2025).
[49]Stefanati, M. et al. Effect of variability of mechanical properties on the predictive capabilities of vulnerable coronary plaques. Comput. Methods Programs Biomed., 254, 108271 (2024).
[50]Song, Z. et al. Synthesizing real-time ultrasound images of muscle based on biomechanical simulation and conditional diffusion network. IEEE Trans. Ultrason. Ferroelectr. Freq. Control, 71(11), 1501–1513 (2024).
[51]Sajjadinia, S. S., Carpentieri, B. & Holzapfel, G. A. Bridging diverse physics and scales of knee cartilage with efficient and augmented graph learning. IEEE Access, 12, 86302–86318 (2024).
[52]Juneja, M. et al. MaxI-Net: A 3D AI framework for CBCT-based maxillofacial defect reconstruction and patient-specific implant generation with biomechanical validation. Bioengineering, 13(6), 619 (2026).
[53]Badrou, A. et al. Human airway material characterization via inverse finite element analysis and neural network surrogate. Biomech. Model. Mechanobiol., 25, 61 (2026).
[54]Ataei, A. et al. The effect of deep learning-based lesion segmentation on failure load calculations of metastatic femurs using finite element analysis. Bone, 179, 116987 (2024).
[55]Alloisio, M. et al. Data Driven Models Merging Geometric, Biomechanical, and Clinical Data to Assess the Rupture of Abdominal Aortic Aneurysms. Eur. J. Vasc. Endovasc. Surg., 70(5), 591–600 (2025).
[56]Eskandari, A. & Sharbatdar, M. Diagnosis of psoriasis and lichen planus in real-time using neural networks based on skin biomechanical properties obtained from numerical simulation. Sci. Rep., 15, 23814 (2025).
[57]Yang, Y. et al. Rapid left ventricle mesh prediction by adaptive deformable model fitting. Phys. Med. Biol., 70(7), 075020 (2025).
[58]Asakura, T. et al. Genetic algorithm-based optimization of columella shape with FEM surrogate modeling: Convergence analysis and application to ossicular chain reconstruction. Biomech. Model. Mechanobiol., 25, 31 (2026).
[59]Xue, C. et al. Multimodal patient-specific registration for breast imaging using biomechanical modeling with reference to AI evaluation of breast tumor change. Life, 11(8), 747 (2021).
[60]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. 國立成功大學機械工程學系碩士論文, 1–83 (2021).
[61]Li, X., Grandvalet, Y. & Davoine, F. Explicit inductive bias for transfer learning with convolutional networks. In Proceedings of the 35th International Conference on Machine Learning, 2825–2834 (2018).
[62]Bozkaya, D. & Müftü, S. Mechanics of the taper integrated screwed-in (TIS) abutments used in dental implants. Journal of Biomechanics, 38(1), 87–97 (2005).
[63]Kennedy, J. & Eberhart, R. Particle swarm optimization. In Proceedings of ICNN'95 - International Conference on Neural Networks, 4, 1942–1948 (1995).