This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
NeRF-based CBCT Reconstruction needs Normalization and Initialization
Created by
Haebom
Author
Zhuowei Xu, Han Li, Dai Sun, Zhicheng Li, Yujia Li, Qingpeng Kong, Zhiwei Cheng, Nassir Navab, and S. Kevin Zhou.
Outline
This paper proposes a study to solve the __T6264_____ of NeRF-based method in Cone Beam Computed Tomography (CBCT) image reconstruction. NeRF-based method causes unstable learning, slow convergence, and poor reconstruction quality due to the local-global learning mismatch problem of hash encoder and neural network. In this paper, we propose a Mapping Consistency Initialization (MCI) strategy that initializes the neural network by utilizing the global mapping property of pre-trained models and a Normalized Hash Encoder (NHEN) that improves feature consistency. The proposed method is simple but effective, and significantly improves the learning efficiency on 128 CT cases collected from four different datasets.
Takeaways, Limitations
•
Takeaways:
◦
We present an effective method to solve the local-global learning mismatch problem in NeRF-based CBCT image reconstruction.
◦
Improved learning stability and convergence speed with regularized hash encoder and mapping consistency initialization strategy.
◦
Achieve performance improvements with simple code modifications.
◦
Validation of effectiveness on a wide range of datasets across various anatomical regions.
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed method.
◦
Lack of comparative analysis with other NeRF-based CBCT reconstruction methods.
◦
Lack of detailed description of the characteristics of the dataset used.
