3DTeethSegX: A Multi-Scenario benchmark for Tooth Point Cloud Completion

Citation Author(s):
lin
zhang
fucheng
niu
Submitted by:
Lin Zhang
Last updated:
Mon, 03/17/2025 - 20:20
DOI:
10.21227/2wev-pb51
License:
0
0 ratings - Please login to submit your rating.

Abstract 

This 3DTeethSegX dataset is a benchmark dataset specifically designed for tooth point cloud completion and segmentation tasks. Built upon the publicly available 3DTeethSeg 2022 MICCAI Challenge dataset, it comprises 1,494 pairs of tooth point clouds and their corresponding tooth images from 38 patients. Each pair includes a partial point cloud (2,048 points) and a complete point cloud (16,384 points). To simulate real-world scenarios, the dataset defines 12 distinct occlusion cases, covering 4 tooth positions (Molar, Premolar, Canine, and Incisor) and 3 levels of occlusion severity (Hard, Moderate, and Simple). The occlusions are generated by clipping spheres at specific locations on the tooth, with sphere radii following a truncated Gaussian distribution tailored to tooth size and occlusion severity. Additionally, the dataset provides simulated occlusal view images, enhancing data diversity and practicality. 3DTeethSegX serves as a high-quality benchmark for tooth point cloud completion, segmentation, and related medical applications.

 

Instructions: 

 Dataset Overview:

3DTeethSegX is a multi-scenario evaluation dataset for tooth point cloud completion tasks. It contains 1,494 pairs of partial and complete point clouds, along with corresponding tooth images. The data is sourced from 38 patients, covering a variety of tooth morphologies and occlusion scenarios.

Dataset Structure:

Partial Point Clouds(in the Partial folder): Each partial point cloud contains 2,048 points, simulating real-world tooth defects.

Complete Point Clouds(in the GT folder): Each complete point cloud contains 16,384 points, serving as ground truth for completion tasks.

Occlusion Cases: The dataset defines 12 occlusion cases based on 4 tooth positions (Molar, Premolar, Canine, Incisor) and 3 severity levels (Hard, Moderate, Simple).

Simulated Images: Occlusal view images are simulated from tooth meshes, mimicking smartphone-captured images for enhanced robustness.

Intended Use Cases:

Tooth Point Cloud Completion: Train and evaluate deep learning models for completing partial tooth point clouds.

Tooth Segmentation: Develop and benchmark algorithms for segmenting teeth from point cloud data.

Potential Research Directions:

Explore advanced point cloud completion and segmentation techniques.

Investigate joint tasks such as completion and segmentation for multi-class tooth point clouds.

Apply the dataset to real-world medical scenarios, such as 3D-printed dental restorations.

Acknowledgments:

We gratefully acknowledge the 3DTeethSeg 2022 MICCAI Challenge dataset for providing the foundational data used to construct the 3DTeethSegX dataset. Their contributions have been instrumental in advancing research in tooth point cloud analysis and related medical applications.