A Multimodal Deep Learning Approach to Predicting Systemic Diseases from Oral Conditions

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Imaging Analysis
• Oral Health and Systemic Disease Linkages

Background:

• Established links between oral diseases (e.g., periodontal disease) and systemic conditions.
• Potential of deep learning, especially with medical imaging, to advance healthcare.
• Value of integrating non-imaging clinical and laboratory data for comprehensive decision-making.

Purpose of the Study:

• To develop a multimodal deep learning approach for predicting systemic diseases from oral health conditions.
• To integrate panoramic radiographs with electronic health record (EHR) data for enhanced prediction accuracy.

Main Methods:

• Utilized a dual-loss autoencoder for feature extraction from 1188 panoramic radiographs.
• Fused extracted image features with demographic and clinical data from EHRs.
• Employed receiver operating characteristic (ROC) curves and accuracy for model evaluation, validated on an unseen dataset.

Main Results:

• The model accurately predicted systemic diseases in Chapters III, VI, and IX with AUC values of 0.92, 0.87, and 0.78, respectively.
• Validation on an unseen dataset showed high accuracy: 0.88 for Chapter III, 0.82 for Chapter VI, and 0.72 for Chapter IX.
• Demonstrated the model's robustness and predictive power across different systemic disease categories.

Conclusions:

• Combining panoramic radiography with clinical oral features via a fusion deep learning model is effective for predicting systemic diseases.
• Highlights the potential of AI in bridging oral health insights with systemic disease diagnostics.
• Suggests a novel approach for early detection and management of systemic conditions through oral health assessment.