3D MODELS OF BONES AND MOTIONS FOR SURGERY AND DIGITAL TWINS

We present a new and general method for creating and maintaining data for models of anatomy and biomechanics and illustrate the method with generative models of the femur for osteotomy planning. The human process of planning surgery and other interventions is essentially a combination of gathering and combining incomplete patient-specific information and filling gaps of knowledge with medical competence and experience from previous cases. Typical examples are 2D medical images of 3D structures, gaps in medical records, or motion capture data only for the lower body. The established medical processes, due to the human expertise, are largely robust towards this situation, but future healthcare will rely more on automation, so automated processes must also be robust to incomplete information and able to learn from experience. A well-known but under-appreciated statistical algorithm of conditional likelihood has the potential to form the backbone of a general system that is robust to incomplete information, in the sense that it will predict any unknown subset of information from any known subset without prior training. To illustrate the potential of the method, we present automatic 3D planning of femur osteotomy, based on input of a small number of desired functional parameters in combination with a previously collected population of femur geometries as illustrated in. In a wider perspective, the algorithm is perfectly suited to work on transaction data, i.e., data received in a continuous stream, such as vital signs from wearable measurement devices and results of ongoing medical examinations. In combination with a model of the physiology, for instance musculoskeletal models, this paves the way for a digital twin as an advanced version of medical records suitable for simulation of the effect of prospective procedures and thereby forming the backbone of shared decision making in healthcare.