Multimodal deep learning for midpalatal suture assessment in maxillary expansion

Area of Science:

• Dentistry and Oral Health
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate midpalatal suture maturation assessment is crucial for effective orthodontic treatment planning.
• Current manual methods suffer from significant inter-examiner variability, impacting treatment reliability.

Purpose of the Study:

• To develop and validate DeepMSM, an automated multimodal deep learning framework for standardized midpalatal suture staging.
• To integrate cone-beam computed tomography (CBCT) with clinical indicators for enhanced diagnostic accuracy.

Main Methods:

• Retrospective analysis of CBCT and lateral cephalometric radiographs from 200 orthodontic patients (aged 7-36 years).
• DeepMSM framework utilized attention-based fusion to integrate multimodal images with clinical variables (age, gender, cervical vertebral maturation stage, mandibular third molar stage).

Main Results:

• DeepMSM achieved 93.75% accuracy and 93.81% F1-score, significantly outperforming single- and dual-modality approaches.
• The system excelled in distinguishing critical stages C and D (92%-93% F1-score), vital for surgical vs. conventional expansion decisions.
• All clinical parameters demonstrated significant correlations with midpalatal suture maturation (p<0.05).

Conclusions:

• DeepMSM offers a novel, highly accurate (93.75%) automated system for midpalatal suture maturation assessment.
• The framework has the potential to reduce diagnostic variability and enhance the reliability of orthodontic treatment decisions.
• This automated tool is particularly beneficial for less experienced clinicians in guiding maxillary expansion therapy.