Results of the summarizing process for the arXiv paper: 2305.18277v1

Teeth localization, segmentation, and labeling from intra-oral 3D scans are crucial tasks in dentistry for improving dental diagnostics, treatment planning, and population-based studies on oral health. To address the challenges associated with developing automated algorithms for teeth analysis due to variations in dental anatomy, imaging protocols, and limited availability of publicly accessible data, the 3DTeethSeg'22 challenge was organized as part of the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2022. The challenge aimed to develop algorithms that can accurately localize, segment, and label teeth from intraoral 3D scans. For this purpose, a dataset containing 1800 scans from 900 patients was prepared with each tooth individually annotated using a human-machine hybrid algorithm. Six algorithms were evaluated on this dataset based on their performance in terms of teeth segmentation accuracy (TSA) and teeth identification rate (TIR). Over five hundred requests for data download and registration were received for the challenge with forty-four teams participating in the preliminary phase and ten teams submitting their final submissions for evaluation. The top six ranked methods were selected based on their performance which included an end-to-end network incorporating graph attentional convolution layers and a global structure branch; a two-stream graph convolutional network (TSGCN); and Dental Arch (DArch), a method introduced for 3D tooth instance segmentation using weak low-cost annotated data. The dataset used in this challenge was acquired from orthodontists/dental surgeons with extensive professional experience following European General Data Protection Regulation (GDPR) agreement from partner dental clinics mainly located in Europe. In summary, the 3DTeethSeg'22 challenge aimed to advance automated teeth analysis by developing algorithms for teeth localization, segmentation, and labeling from intraoral 3D scans.

• Teeth localization, segmentation, and labeling from intra-oral 3D scans are crucial tasks in dentistry.
• The 3DTeethSeg'22 challenge was organized as part of the MICCAI conference in 2022.
• The challenge aimed to develop algorithms for accurately localizing, segmenting, and labeling teeth from intraoral 3D scans.
• A dataset containing 1800 scans from 900 patients was prepared with each tooth individually annotated using a human-machine hybrid algorithm.
• Six algorithms were evaluated based on their teeth segmentation accuracy (TSA) and teeth identification rate (TIR).
• Over five hundred requests for data download and registration were received for the challenge.
• Forty-four teams participated in the preliminary phase and ten teams submitted final submissions for evaluation.
• The top six ranked methods included an end-to-end network incorporating graph attentional convolution layers, a two-stream graph convolutional network (TSGCN), and Dental Arch (DArch).
• The dataset used in the challenge was acquired following GDPR agreement from partner dental clinics mainly located in Europe.

The 3DTeethSeg'22 Challenge: Advancing Automated Teeth Analysis

In dentistry, teeth localization, segmentation, and labeling from intra-oral 3D scans are crucial tasks for improving dental diagnostics, treatment planning, and population-based studies on oral health. However, the development of automated algorithms for teeth analysis is challenging due to variations in dental anatomy, imaging protocols, and limited availability of publicly accessible data. To address these challenges and advance automated teeth analysis, the 3DTeethSeg'22 challenge was organized as part of the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) in 2022.

Objective & Dataset

The objective of this challenge was to develop algorithms that can accurately localize, segment, and label teeth from intraoral 3D scans. For this purpose a dataset containing 1800 scans from 900 patients was prepared with each tooth individually annotated using a human-machine hybrid algorithm. The dataset used in this challenge was acquired from orthodontists/dental surgeons with extensive professional experience following European General Data Protection Regulation (GDPR) agreement from partner dental clinics mainly located in Europe.

Participants & Evaluation Metrics

Over five hundred requests for data download and registration were received for the challenge with forty-four teams participating in the preliminary phase and ten teams submitting their final submissions for evaluation. The six algorithms evaluated on this dataset were based on their performance in terms of two metrics: teeth segmentation accuracy (TSA) and teeth identification rate (TIR). The top six ranked methods were selected based on their performance which included an end-to-end network incorporating graph attentional convolution layers and a global structure branch; a two-stream graph convolutional network (TSGCN); Dental Arch (DArch), a method introduced for 3D tooth instance segmentation using weak low-cost annotated data; among others.

Conclusion

In summary, the 3DTeethSeg'22 challenge aimed to advance automated teeth analysis by developing algorithms for teeth localization, segmentation, and labeling from intraoral 3D scans. This initiative brought together researchers across disciplines to develop innovative solutions to improve dental diagnostics through automated tools that can accurately identify individual teeth within an intraoral scan image.