A rigged model of the breast for preoperative surgical planning.

In breast surgical practice, various scans and medical examinations are performed before surgery. This includes identifying landmarks defining the operating procedure. In most cases, the position of the patient during the scan is vastly different from the one encountered during the operation. We address the challenge of mapping preoperative information to the operating field, with the following constraints: registration has to be done in less than 10 seconds to be compatible with a clinical workflow; the cost of the device must be small and we assume data scarcity, i.e. that our database has twenty scans of patients at most. We build anatomical complexity through a skinning model comprised of scalable bones (to account for pose and morphological variations) and deformable organs (blendshapes, to account for anatomical variations). Similar to animation rigs used in computer graphics, and in contrast to statistical approaches, we manually design a model with some desirable properties, using a reduced number of well-chosen degrees of freedom. Meaningful constraints can be applied to the registration depending on the context, and the trade-off between precision and complexity can be optimized. The result is a surface mesh of the patient obtained in less than 1 minute (scan and reconstruction included) and a registration method that converges within a few seconds (3 maximum), reaching a mean absolute squared error of 2.3 mm for mesh registration and 8.0 mm for anatomical landmarks. The registered model is used to transfer surgical reference patterns on any patient in any position.