Clinical Report: Two-Phase Deep Learning Approach for Diagnosing Pediatric OSA
Overview
This study presents a two-phase deep learning framework for diagnosing pediatric obstructive sleep apnea-hypopnea syndrome (OSAHS) using lateral cephalometric images. The model demonstrated high accuracy in segmentation and classification, significantly improving diagnostic performance among dentists.
Background
Pediatric OSAHS is a common disorder affecting 1%–4% of children, linked to serious comorbidities and increased healthcare utilization. Traditional diagnostic methods like polysomnography are costly and not widely accessible, highlighting the need for efficient screening tools. Lateral cephalograms, routinely used in dental practices, could serve as a valuable resource for opportunistic screening if enhanced by artificial intelligence.
Data Highlights
Metric
Value
Confidence Interval
Upper airway segmentation DSC
0.931
N/A
Upper airway segmentation IoU
0.872
N/A
Fusion model AUC
0.945
(0.863–0.994)
Fusion model F1 score
0.933
(0.818–0.995)
LCs model AUC
0.797
(0.585–0.968)
Mask-based ROI model AUC
0.882
(0.748–0.983)
Key Findings
The two-phase framework achieved a mean DSC of 0.931 for upper airway segmentation.
The fusion model for OSAHS classification reached an AUC of 0.945, outperforming traditional models.
Grad-CAM highlighted anatomically relevant areas, enhancing model interpretability.
AI assistance improved diagnostic accuracy by 0.165 for junior dentists and 0.237 for senior dentists.
This model could facilitate early detection and individualized management of pediatric OSAHS.
Clinical Implications
The proposed deep learning framework offers a promising alternative for diagnosing pediatric OSAHS using routine dental imaging. Its high accuracy and interpretability may support clinicians in making timely and informed decisions regarding patient management.
Conclusion
This study underscores the potential of integrating artificial intelligence into routine dental practices for the early diagnosis of pediatric OSAHS, which could lead to improved patient outcomes.
Colorectal cancer (CRC) impacts about 1 in 13,000 pregnancies in the United States. This article follows the case of a 36-year-old woman diagnosed with CRC while pregnant and highlights considerations around CRC diagnosis and cancer treatment during pregnancy.
