Automatic Multi-structure Segmentation in Cone Beam Computed Tomography Volumes Using Deep Encoder-Decoder Architectures
| Type of publication: | Article |
| Citation: | Wodzinski, M., Müller, H. (2025). Automatic Multi-structure Segmentation in Cone Beam Computed Tomography Volumes Using Deep Encoder-Decoder Architectures. In: Wang, Y., et al. Supervised and Semi-supervised Multi-structure Segmentation and Landmark Detec |
| Publication status: | Published |
| Journal: | Supervised and Semi-supervised Multi-structure Segmentation and Landmark Detection in Dental Data (MICCAI 2024) |
| Volume: | 15571 |
| Year: | 2025 |
| Month: | May |
| Pages: | 63-71 |
| ISSN: | 978-3-031-88976-9 |
| URL: | https://link.springer.com/chap... |
| DOI: | https://doi.org/10.1007/978-3-031-88977-6_7 |
| Abstract: | Automatic multi-structure segmentation in cone beam computed tomography is an important area of research in dentistry. Accurate segmentation of structures such as the mandible, maxillary bones, pharynx, and inferior alveolar canal is crucial for all head and neck surgical specialties. The importance of the task motivated the researchers to organize a dedicated challenge, ToothFairy2, aiming to benchmark distinct segmentation methods from researchers from all over the world. In this work, we present our contribution to the challenge, based on a patch-based encoder-decoder segmentation architecture. We compare several building blocks and show that, for this particular task, the most important factor is a data sampling strategy that correctly balances the class distribution. During the challenge, our method achieved a considerable score for the lower jaw structures (avgDSC = 0.878) but did not adequately segment the upper jaw structures due to incorrect case-level sampling (avgDSC = 0.738). |
| Keywords: | Cone Beam Computed Tomography Deep Learning Image Segmentation ToothFairy |
| Authors | |
| Added by: | [] |
| Total mark: | 0 |
|
Attachments
|
|
|
Notes
|
|
|
|
|
|
Topics
|
|
|
|
|