Welcome to the ToothFairy3 Challenge 🧚♀️¶
The ToothFairy3 challenge is part of ODIN2025 (Oral and Dental Image aNalysis challenges) cluster of challenges, which will be hosted by the homonymous ODIN2025 workshop @ MICCAI2025. The challenge involves two different tasks, namely Task 2 and Task 3 on the Structured Submission Document!
Background¶
The use of Cone Beam Computed Tomography (CBCT) is rapidly expanding in dentistry, head and neck, and maxillofacial surgery due to its short acquisition time, low radiation dose, and high resolution of hard tissues. In this context, the ToothFairy challenge series has progressively advanced the state-of-the-art in CBCT segmentation. Starting with ToothFairy - MICCAI 2023, the focus was on the segmentation of the Inferior Alveolar Canal (IAC), a critical structure within the mandible whose accurate identification is vital for surgical planning and risk minimization. ToothFairy2 - MICCAI 2024 extended this initiative to include a broader set of anatomical structures—such as the mandible, teeth, maxillary bone, and pharynx—resulting in a comprehensive 42-class segmentation task. These structures are of cross-disciplinary importance, spanning surgical, clinical, and anesthesiological applications. Deep learning-based automatic segmentation models have shown encouraging results, demonstrating the potential to support medical personnel in routine workflows.
With ToothFairy3 - MICCAI 2025, we aim to further advance this effort in two key directions: first, by expanding the dataset with more publicly available, 3D-annotated CBCT scans and introducing new anatomical classes—i.e., pulp cavities, incisive nerves, and the lingual foramen for a total of 77 classes—which are especially relevant for orthodontic procedures. Second, we emphasize the importance of computational efficiency by incorporating runtime as a primary evaluation metric alongside accuracy, reflecting the real-world demand for fast and reliable tools in clinical practice (Task 1).
In parallel, this edition introduces a novel interactive segmentation task focused on the Inferior Alveolar Canal (IAC). Despite progress in automated methods, the IAC remains a challenging structure to segment due to its fine-grained and variable nature. To bridge the gap between automation and clinical applicability, we propose a click-based interactive segmentation track, allowing users to guide the model with minimal input while significantly enhancing precision. Participants are encouraged to explore emerging prompt-based foundation models and design innovative interactive methods tailored for CBCT imagery (Task 2).
ToothFairy3 thus aims to foster both automated and interactive solutions, ultimately advancing the integration of AI-driven segmentation tools into daily clinical workflows and supporting improved patient outcomes.
Resources¶
Getting started:
Task 1 baseline template and evaluation code available soon
Task 2 baseline template and evaluation code available soon
Public datasets for local development: check our Dataset page.
Announcements¶
📰 09/05/2025: The first version of the ToothFairy3 dataset has been officially released. Visit the Dataset page! 🚀