Medical AI & 3D Reconstruction
Medical AI describes the area of artificial intelligence that deals with machine learning and deep learning pipelines that elevate and augment the effectiveness and efficiency of processes in the medical field. Computer graphics, in this context, is mostly used for simulations of the human body in clinical scenarios. Clinically accurate simulations require the respective 3D models to replicate the human anatomy with high accuracy. However, while measurements of the outer part of the human body can safely be taken without any risks, accurately capturing the internal structures, like organs or bones, often requires highly invasive x-ray imaging.
Our project, "AutoSkull", was developed to specifically tackle the problem of predicting the shape of the human skull with acceptable accuracy, without having to use invasive x-ray technology. The method uses a deep learning model that operates on a latent feature vector inside a PCA-space of 3D shapes. This latent feature vector is the input to predict another point inside of the same PCA space that can be used to recover a prediction of a 3D skull model, given a 3D head model as input.
This project was done in collaboration with the ETH Computer Graphics Laboratory and an industry partner. Our paper was published at the MICCAI 2024 conference, where I was in attendance on behalf of this project. Additionally, the method of this approach was patented by our industry partner.