3d reconstruction for plastic surgery simulation based on statistical shape models
- Autores: Guillermo Ruiz Fernández
- Directores de la Tesis: Miguel Ángel González Ballester (dir. tes.), Federico Mateo Sukno (codir. tes.)
- Lectura: En la Universitat Pompeu Fabra ( España ) en 2018
- Idioma: español
- Tribunal Calificador de la Tesis: Kostantyn Butakov (presid.), Xavier Binefa i Valls (secret.), Javier Herrero Jover (voc.)
- Programa de doctorado: Programa de Doctorado en Tecnologías de la Información y las Comunicaciones por la Universidad Pompeu Fabra
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
This thesis has been accomplished in Crisalix in collaboration with the Universitat Pompeu Fabra within the program of Doctorats Industrials. Crisalix has the mission of enhancing the communication between professionals of plastic surgery and patients by providing a solution to the most common question during the surgery planning process of “How will I look after the surgery?”. The solution proposed by Crisalix is based in 3D imaging technology. This technology generates the 3D reconstruction that accurately represents the area of the patient that is going to be operated. This is followed by the possibility of creating multiple simulations of the plastic procedure, which results in the representation of the possible outcomes of the surgery.
This thesis presents a framework capable to reconstruct 3D shapes of faces and breasts of plastic surgery patients from 2D images and 3D scans. The 3D reconstruction of an object is a challenging problem with many inherent ambiguities. Statistical model based methods are a powerful approach to overcome some of these ambiguities. We follow the intuition of maximizing the use of available prior information by introducing it into statistical model based methods to enhance their properties.
First, we explore Active Shape Models (ASM) which are a well-known method to perform 2D shapes alignment. However, it is challenging to maintain prior information (e.g. small set of given landmarks) unchanged once the statistical model constraints are applied. We propose a new weighted regularized projection into the parameter space which allows us to obtain shapes that, at the same time, fulfill the imposed shape constraints and are plausible according to the statistical model.
Second, we extend this methodology to be applied to 3D Morphable Models (3DMM), which are a widespread method to perform 3D reconstruction. However, existing methods present some limitations. Some of them are based in non-linear optimizations computationally expensive that can get stuck in local minima. Another limitation is that not all the methods provide enough resolution to represent accurately the anatomy details needed for this application. Given the medical use of the application, the accuracy and robustness of the method, are important factors to take into consideration. We show how 3DMM initialization and 3DMM fitting can be improved using our weighted regularized projection.
Finally, we present a framework capable to reconstruct 3D shapes of plastic surgery patients from two possible inputs: 2D images and 3D scans. Our method is used in different stages of the 3D reconstruction pipeline: shape alignment; 3DMM initialization and 3DMM fitting. The developed methods have been integrated in the production environment of Crisalix, proving their validity.
