Toward Personalized Dental Surgery Planning: Dental Bone Stiffness Discovery and Strain Prediction Using Physics-Informed Machine Learning - Project Summary Dental implants significantly enhance quality of life by effectively replacing missing teeth; however, mechan- ical loosening and implant failure frequently compromise treatment outcomes. The success of dental implants critically depends on achieving the optimal strain levels in the bones that interface with the implants. Insufficient strain can lead to bone loss due to disuse, eventually resulting in implant loosening and failure, whereas exces- sive strain may cause bone fractures. Bone stiffness varies significantly with age and sex, between individuals, and even within the same individual. This distributed, non-uniform stiffness affects the strain levels under applied loads, significantly influencing implant biomechanics and the resulting bone healing. Surgical decisions, such as implant stiffness, diameter, and placement location, can also impact the strain distribution in the bone-implant complex. During the surgery planning process, precise and rapid assessment of bone stiffness and strain is essential for personalized surgical planning and implant design. However, conventional techniques for measuring bone stiffness are either invasive or ill-suited for detailed 3D internal analysis. Furthermore, current methods for predicting strain are computationally expensive, thus falling short of clinical applicability. This R21 project proposes an innovative strategy for personalized dental implant planning through precise and rapid prediction of bone stiffness and strain, employing biomechanical physics-informed machine learning (PIML). This approach leverages the unprecedented opportunities presented by the availability of computed tomography (CT) imaging data coupled with mechanical testing, combined with machine learning (ML), to offer a detailed and individualized evaluation of bone quality. By leveraging the principles of biomechanical physics, the proposed ML methodology will infer the underlying bone properties, facilitating efficient predictions based on individual CT images and clin- ical factors. The project is organized into two primary goals: Establishing ML Methods for Personalized 3D Bone Stiffness Estimation (Specific Aim 1): By leveraging recent advancements in 3D deformation data ac- quisition and integrating these with the linear elasticity model, the proposed method aims to derive precise bone stiffness measurements from CT images, incorporating factors such as age and sex to reflect individual physio- logical differences essential for successful implant integration. Establishing ML Methods for Strain Prediction in Response to Surgical Decisions (Specific Aim 2): Sophisticated ML techniques, including U-net architec- ture and few-shot learning, will be employed to establish an effective predictive model that considers individuals’ unique bone geometries, thereby enhancing treatment precision and effectiveness. By merging data-driven tech- niques with biomechanical physics, this project is set to substantially improve the accuracy and safety of dental implants. We anticipate a paradigm shift towards more effective, customized dental care, with potential appli- cations extending into orthopedic surgery. This research highlights the crucial role of integrating technological advances with medical practices to elevate health outcomes.
