Supplementary file 1_Interpretable two-stage deep learning for pediatric obstructive sleep

188 lateral cephalogram images from children at two hospitals between 2021 and 2025 were used to develop an AI diagnostic framework. The proposed two-stage model segments the upper airway and classifies obstructive sleep apnea, achieving a mean Dice Similarity Coefficient of 0.931 for segmentation and an AUC of 0.945 for classification. Author Jiayi Zhang released this research dataset on figshare under a CC-BY-4.0 license.

Use Cases

• Training image segmentation models for upper airway delineation based on lateral cephalograms mentioned in the description.
• Developing interpretable classification models for pediatric obstructive sleep apnea using craniofacial and airway features.
• Benchmarking AI-assisted diagnostic tools against human performance in a clinical reader study context.
• Studying model interpretability techniques like Grad-CAM on medical imaging tasks.

Strengths

• Model performance is quantitatively validated with a mean DSC of 0.931 for segmentation and an AUC of 0.945 for classification.
• Clinical utility is demonstrated via a reader study showing AI assistance increased diagnostic accuracy for dentists.
• The dataset is derived from a multi-hospital study involving 188 pediatric patients, providing a clinical foundation.

Limitations

• The dataset size is 3.9 MB, which is small for training complex deep learning models from scratch.
• Row count and column-level documentation are absent; field semantics must be inferred after download.
• Data may reflect temporal and institutional bias inherent to the two hospitals used for collection.

Provenance

Source

figshare, author Jiayi Zhang.

Collection Method

Retrospective collection of lateral cephalograms from two hospitals.

Time Range

January 2021 to October 2025.

Freshness

Last updated 2026-04-21 05:23:45; freshness should be verified.