Building on the voxel-wise segmentation benchmarks established by ToothFairy 1-3, ToothFairy4 (ODIN 2026 Track 1) shifts the focus from geometric output to automatic clinical report generation. The goal is to move beyond 3D anatomy mapping toward 3D-to-text models that can synthesize surgical planning reports directly from CBCT volumes.

Challenge Focus

• Clinical Reasoning. Systems must evaluate dental status, bone quality, and anatomical risks to generate structured reports for procedures like implant placement and sinus lifts.

• Multimodal Integration. Developing architectures that translate high-dimensional 3D imaging into coherent clinical documentation.

• Generalization. Testing robustness across multi-center data and diverse acquisition protocols to ensure clinical readiness.

ToothFairy4 aims to standardize documentation and accelerate surgical workflows, transitioning from pure segmentation to scalable multimodal clinical decision support.

Over the past years, the ODIN challenge series (including 3DTeethSeg/Land challenges) has established large-scale benchmarks for tooth segmentation and landmark detection from 3D intraoral scans. Building on this foundation, ODIN 2026 Track 2 (Bite2Text) represents the next step toward end-to-end clinical understanding of intraoral data. Given the challenges of report preparation, such as time consumption, manual workload, and inter-observer variability, this track focuses on the automatic generation of orthodontic reports from multimodal inputs, combining 3D intraoral scans and 2D intraoral photographs. It aims to foster the development and benchmarking of systems that produce structured clinical reports to improve safety and consistency in clinical practice. Participants are tasked with developing multimodal 2D/3D-to-text models that jointly reason over dental geometry and visual appearance to produce clinically meaningful reports. The dataset includes acquisitions from multiple centers, scanners, and clinical settings, with evaluation conducted on a hidden test set collected from a center not included in the training data to assess robustness and generalization. By moving beyond intermediate geometric tasks, this challenge aims to foster clinically deployable decision-support systems for orthodontic practice, targeting generalization across scanners, protocols, and patient populations.

Building on the STSR challenge series from ODIN 2025, STS 2026 expands the scope from pure segmentation toward a comprehensive multimodal dental intelligence benchmark, integrating artifact-robust perception, cross-modal alignment, and clinical language generation.

Over the past editions, ODIN has established rigorous benchmarks for tooth segmentation, landmark detection, and reproducible 3D analysis. STS 2026 advances this trajectory by addressing three interdependent bottlenecks that hinder end-to-end clinical deployment: (1) metal-induced artifacts degrading CBCT interpretability, (2) modality gaps impeding high-fidelity CBCT-IOS fusion, and (3) the lack of automated, structured reporting from 3D dental imaging. This challenge provides curated, multi-center datasets and transparent evaluation protocols to foster methods that are not only accurate but also clinically actionable, generalizable, and ready for real-world integration.

Challenge Focus

• Artifact-Robust Perception. Develop unified frameworks that jointly suppress metal streak artifacts in CBCT and achieve high-precision segmentation of teeth and complex root canal systems under anatomical variability and low-contrast conditions.
• Cross-Modal Registration. Design geometry-aware, feature-consistent algorithms to accurately align CBCT (internal hard-tissue architecture) with intraoral scans (external crown morphology), enabling integrated “internal-external” digital modeling for surgical planning.
• Clinical Report Generation. Leverage the MMDental dataset, the large-scale public resource pairing 3D CBCT volumes with expert-reviewed medical records, to train multimodal 3D-to-text models that generate structured, clinically faithful diagnostic reports directly from imaging data.

STS 2026 aims to bridge the gap between algorithmic innovation and clinical utility in digital dentistry. By standardizing evaluation across perception, alignment, and reasoning tasks, this challenge accelerates the development of reproducible, end-to-end AI systems that can support preoperative simulation, intraoperative guidance, and postoperative assessment in precision oral healthcare.