Deep learning-based automatic adenoid segmentation and a novel volume-based index for adenoid hypertrophy assessment

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Sleep Medicine

Background:

• Adenoid hypertrophy is a primary cause of pediatric obstructive sleep apnea (OSA), impacting cognitive and craniofacial development.
• Accurate assessment of adenoid volume and nasopharyngeal obstruction is crucial for timely intervention.
• Current methods for quantifying airway obstruction lack precision in three dimensions.

Purpose of the Study:

• To develop a deep learning-based method for accurate adenoid segmentation from CBCT scans.
• To establish a novel quantitative index, the Three-Dimensional Adenoidal-Nasopharyngeal Ratio (3D-AN), for assessing nasopharyngeal airway obstruction.
• To investigate the correlation between the 3D-AN and pediatric OSA.

Main Methods:

• A SegResNet-based deep learning model was trained on CBCT scans from pediatric OSA patients to segment adenoids.
• Adenoid volume was determined by predicting postoperative airway morphology from preoperative scans.
• The 3D-AN ratio was calculated using adenoid and nasopharyngeal volumes, and its correlation with OSA was analyzed using polysomnography data.

Main Results:

• The adenoid segmentation model achieved a Dice similarity coefficient of 0.88 and a relative volume error of 0.09.
• A 3D-AN ratio below 0.18 showed no significant correlation with pediatric OSA.
• A 3D-AN ratio above 0.18 demonstrated a significant correlation with pediatric OSA (r=0.56, P<0.01).

Conclusions:

• An automated deep learning method provides accurate adenoid segmentation from CBCT.
• The novel 3D-AN ratio shows potential for early detection of adenoid-related airway obstruction.
• This approach can support future research and clinical management of pediatric airway obstruction.