2021/09/10 by Zachary M. C. Baum, Yipeng Hu, Baum, Zachary M C +3
Computer Science · Engineering · Medicine · #3D Shape Modeling and Analysis #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Medical Image Segmentation Techniques #Prostate Cancer Diagnosis and Treatment #Robotics and Sensor-Based Localization #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2109.05023
openalex publication_date 2021/09/10 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
We present Free Point Transformer (FPT) - a deep neural network architecture\nfor non-rigid point-set registration. Consisting of two modules, a global\nfeature extraction module and a point transformation module, FPT does not\nassume explicit constraints based on point vicinity, thereby overcoming a\ncommon requirement of previous learning-based point-set registration methods.\nFPT is designed to accept unordered and unstructured point-sets with a variable\nnumber of points and uses a "model-free" approach without heuristic\nconstraints. Training FPT is flexible and involves minimizing an intuitive\nunsupervised loss function, but supervised, semi-supervised, and partially- or\nweakly-supervised training are also supported. This flexibility makes FPT\namenable to multimodal image registration problems where the ground-truth\ndeformations are difficult or impossible to measure. In this paper, we\ndemonstrate the application of FPT to non-rigid registration of prostate\nmagnetic resonance (MR) imaging and sparsely-sampled transrectal ultrasound\n(TRUS) images. The registration errors were 4.71 mm and 4.81 mm for complete\nTRUS imaging and sparsely-sampled TRUS imaging, respectively. The results\nindicate superior accuracy to the alternative rigid and non-rigid registration\nalgorithms tested and substantially lower computation time. The rapid inference\npossible with FPT makes it particularly suitable for applications where\nreal-time registration is beneficial.\n