Deep learning on lateral cephalograms estimates mid-palatal suture maturity to guide expander selection

Deep learning models using lateral cephalograms estimated mid-palatal suture maturity with AUCs of 0.91–0.97 on external validation, potentially avoiding CBCT when ionizing radiation is not justified, according to a study of 543 paired scans.

• Models incorporating the TKSGA attention module demonstrated better discriminative ability, calibration, and net clinical benefit than baseline architectures.
• Heatmap analysis showed the deep learning models focused on morphological features of the dentition, cervical vertebrae, and nasopharyngeal airway.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.