MICCAI STS 2024 Challenge: Semi-Supervised Instance-Level Tooth Segmentation in Panoramic X-ray and CBCT Images
🤖 AI Summary
To address the scarcity of labeled training data for instance-level tooth segmentation in dental imaging—caused by high annotation costs—this study proposes a hybrid framework integrating foundation models with semi-supervised learning (SSL). Methodologically, we conduct the first systematic evaluation of SSL paradigms on large-scale 2D orthopantomograms (OPGs) and 3D cone-beam computed tomography (CBCT) dental datasets, incorporating pseudo-labeling, consistency regularization, knowledge distillation, and a SAM-guided coarse-to-fine segmentation pipeline. Our key contributions include the novel integration of vision foundation models into dental semi-supervised instance segmentation and the design of a multi-stage refinement strategy. Experiments on the MICCAI STS 2024 challenge benchmark demonstrate that our best-performing configuration improves instance affinity (IA) by 44.2 percentage points on 2D OPGs and instance Dice score by 61.1 percentage points on 3D CBCT volumes—substantially outperforming fully supervised baselines. This work establishes a new paradigm for high-accuracy tooth segmentation under low-labeling-cost constraints.
Application Category
📝 Abstract
Orthopantomogram (OPGs) and Cone-Beam Computed Tomography (CBCT) are vital for dentistry, but creating large datasets for automated tooth segmentation is hindered by the labor-intensive process of manual instance-level annotation. This research aimed to benchmark and advance semi-supervised learning (SSL) as a solution for this data scarcity problem. We organized the 2nd Semi-supervised Teeth Segmentation (STS 2024) Challenge at MICCAI 2024. We provided a large-scale dataset comprising over 90,000 2D images and 3D axial slices, which includes 2,380 OPG images and 330 CBCT scans, all featuring detailed instance-level FDI annotations on part of the data. The challenge attracted 114 (OPG) and 106 (CBCT) registered teams. To ensure algorithmic excellence and full transparency, we rigorously evaluated the valid, open-source submissions from the top 10 (OPG) and top 5 (CBCT) teams, respectively. All successful submissions were deep learning-based SSL methods. The winning semi-supervised models demonstrated impressive performance gains over a fully-supervised nnU-Net baseline trained only on the labeled data. For the 2D OPG track, the top method improved the Instance Affinity (IA) score by over 44 percentage points. For the 3D CBCT track, the winning approach boosted the Instance Dice score by 61 percentage points. This challenge confirms the substantial benefit of SSL for complex, instance-level medical image segmentation tasks where labeled data is scarce. The most effective approaches consistently leveraged hybrid semi-supervised frameworks that combined knowledge from foundational models like SAM with multi-stage, coarse-to-fine refinement pipelines. Both the challenge dataset and the participants' submitted code have been made publicly available on GitHub (https://github.com/ricoleehduu/STS-Challenge-2024), ensuring transparency and reproducibility.
Problem
Research questions and friction points this paper is trying to address.
Addresses data scarcity in tooth segmentation via semi-supervised learning
Benchmarks SSL methods for instance-level segmentation in dental images
Evaluates performance gains of SSL over fully-supervised baselines
Innovation
Methods, ideas, or system contributions that make the work stand out.
Semi-supervised learning addresses data scarcity in tooth segmentation
Hybrid frameworks combine foundational models with multi-stage refinement
Public dataset and code ensure transparency and reproducibility
🔎 Similar Papers
No similar papers found.
