Data supporting the dissertation "Towards Reliable and Generalizable Cardiac MRI Segmentation: Probabilistic and Imaging Physics Perspectives".
Year: 2026
Type: dataset
Abstract: Dataset supporting Yidong Zhao's PhD dissertation, “Towards Reliable and Generalizable Cardiac MRI Segmentation: Probabilistic and Imaging Physics Perspectives.” The research investigates how to improve the reliability and generalizability of AI-based cardiac magnetic resonance (CMR) image segmentation. In particular, it explores probabilistic approaches for characterizing segmentation uncertainty and imaging-physics-based approaches for improving generalization across different CMR image contrasts and acquisition settings. This dataset contains trained neural network weights generated during the PhD research. The included models cover several components of the dissertation, including Hamiltonian Monte Carlo-based probabilistic segmentation, probabilistic contour prediction, right ventricular segmentation, and the Reverse Imaging framework for contrast-generalizable CMR segmentation. The weights were obtained by training the corresponding deep learning models on cardiac MR imaging datasets used in the associated studies. The dataset contains trained model parameters rather than raw medical images. When used together with the accompanying source code, these weights enable inference on compatible cardiac MR images and support reproduction and further evaluation of the methods described in the dissertation.The corresponding implementation and instructions for using the trained weights are available in the https://gitlab.tudelft.nl/yidongzhao/yd-phd-code-guidance.
Source: 4TU.ResearchData
Author: Yidong Zhao
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204435972