Deep learning model for differentiating thyroid eye disease and orbital myositis on computed tomography (CT) imaging

Area of Science:

• Ophthalmology
• Radiology
• Artificial Intelligence

Background:

• Thyroid eye disease (TED) and orbital myositis present similar clinical symptoms, complicating differential diagnosis.
• Accurate differentiation is crucial for appropriate patient management and treatment strategies.

Purpose of the Study:

• To develop and validate a deep learning model for distinguishing between TED and orbital myositis using orbital computed tomography (CT) imaging.
• To assess the diagnostic performance of the AI model in differentiating these two orbital inflammatory conditions.

Main Methods:

• A retrospective cohort study included 192 patients (110 TED, 51 orbital myositis, 31 controls) over 12 years.
• A Visual Geometry Group-16 deep learning network was trained on coronal orbital CT slices.
• The model was trained using binary combinations of TED, orbital myositis, and control groups.

Main Results:

• The model achieved 98.4% accuracy and an AUC of 0.999 in differentiating orbital myositis from TED.
• For orbital myositis detection, the model demonstrated a sensitivity of 0.964, specificity of 0.994, and F1 score of 0.984.
• A total of 1628 orbital CT images were analyzed.

Conclusions:

• Deep learning models can accurately differentiate TED and orbital myositis from single coronal orbital CT images.
• The model's ability to identify features beyond muscle enlargement suggests broader diagnostic potential.
• This AI-driven approach offers a promising tool for improving the diagnosis of orbital inflammatory diseases.