BronchoAccess Open Dataset

Oct 12, 2026

Can a chest CT help predict whether bronchoscopy will yield a diagnostic biopsy—before the procedure begins?

Diagram of local data, local model training, and shared parameter updates. — Conceptual research schematic.

We are introducing BronchoAccess, our distinctive dataset and benchmark for outcome-driven bronchoscopy accessibility prediction, presented in our MICCAI 2026** work.

What makes this dataset special is its connection between pre-procedural imaging and actual clinical outcomes:

🔹 438 patients across three clinical centers

🔹 3D CT scans linked to documented bronchoscopy outcomes

🔹 279 diagnostic and 159 nondiagnostic procedures, with successful outcomes defined by histopathological confirmation of diagnostic tissue

These data open opportunities to study how lesion appearance, anatomical location, and airway geometry together inform procedural planning.

Alongside the dataset, we introduce an anatomy-aware Mixture-of-Experts model that combines local CT features, lung-lobe information, and airway-path geometry. It achieved an AUROC of 0.805 in patient-level five-fold cross-validation, establishing a benchmark for further research.

We invite the medical imaging, pulmonary medicine, and interventional AI communities to explore BronchoAccess, develop new methods, and help advance outcome-informed planning.

🤗 Dataset: [BronchoAccess on Hugging Face](https://huggingface.co/datasets/linkai-peng/BronchoAccess/)

🌐 Project, paper, and citation: [BronchoAccess project page](https://nubagcilab.github.io/BronchoAccess/)

📖 If you use BronchoAccess in your research, please cite our MICCAI 2026 paper:

Peng et al. “Anatomy-Aware Prediction of Bronchoscopic Accessibility from 3D CT.” MICCAI, 2026.

Thanks to our entire multidisciplinary team for bringing this work together!

#MICCAI2026 #MedicalImaging #OpenData #Bronchoscopy #AIinHealthcare