This master's thesis presents a method for the automatic segmentation of uteri in 3D ultrasound data and a reference model of a healthy uterus. The work is part of the NURSE research, which aims to determine the criteria for a normal uterus and identify possible deviations in women with infertility and repeated miscarriages. In this work, we present a deep neural model for the automatic segmentation of the uterus, which achieved a Dice score of 0.899, an algorithm for the alignment of segmented 3D shapes, and visualization of the average model of the uterus obtained with the presented methods. We will also publish a public database of annotated volumetric ultrasound data. The proposed methods and findings provide valuable insights into the analysis of uterine shape and contribute to the field of gynecology.
|
