Deep Learning-Driven Exophthalmometry through Facial Photographs in Thyroid Eye Disease

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Thyroid eye disease (TED) often causes proptosis, requiring accurate measurement.
• Traditional exophthalmometry methods can be challenging and require specialized equipment.

Purpose of the Study:

• To develop and evaluate a deep learning (DL)-assisted system for proptosis measurement using facial photographs in TED patients.
• To assess the accuracy and reliability of this novel system compared to standard measurements.

Main Methods:

• A retrospective cohort study involving 1108 TED patients from Severance Hospital (SH) and 171 from Seoul National University Bundang Hospital (SNUBH).
• A dual-stream ResNet-18 neural network trained on facial images and depth maps, validated against Hertel exophthalmometry measurements.
• Accuracy assessed using Mean Absolute Error (MAE), Pearson correlation, Intraclass Correlation Coefficient (ICC), and Receiver Operating Characteristic (ROC) curve analysis.

Main Results:

• The DL system achieved low MAE (1.27 mm for SH, 1.24 mm for SNUBH) and strong correlations (Pearson: 0.82 SH, 0.77 SNUBH; ICC: 0.80 SH, 0.73 SNUBH).
• ROC analysis demonstrated high accuracy in detecting proptosis (AUC: 0.91 SH, 0.88 SNUBH).
• The system accurately detected significant proptosis changes (≥ 2 mm) with 74.6% accuracy.

Conclusions:

• The DL-assisted system provides an accurate and accessible method for exophthalmometry in TED patients using facial photographs.
• This technology offers a promising alternative to traditional methods, potentially improving proptosis measurement accessibility in various clinical settings.