A Deep Learning Model for Screening Computed Tomography Imaging for Thyroid Eye Disease and Compressive Optic...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Thyroid Eye Disease (TED) is an autoimmune condition that can lead to vision loss through compressive optic neuropathy.
• Early detection and monitoring of TED are crucial to prevent permanent visual impairment.
• Deep learning (AI) shows promise in medical image analysis for disease detection.

Purpose of the Study:

• To develop and evaluate a deep learning model for identifying TED and compressive optic neuropathy using orbital CT scans.
• To compare the AI model's diagnostic accuracy against human expert performance.

Main Methods:

• Retrospective analysis of 1187 orbital CT scans from 141 patients (TED and controls).
• A deep learning model (Visual Geometry Group-16) was trained to classify no TED, mild TED, and severe TED with optic neuropathy.
• Model performance was evaluated and compared to an oculoplastic surgeon's assessment.

Main Results:

• The deep learning model achieved 89.5% accuracy in distinguishing between no TED, mild TED, and severe TED with optic neuropathy.
• The AI model demonstrated a superior area under the curve (0.96-0.99) compared to the oculoplastic surgeon's accuracy (70.0%).
• The model accurately identified TED and TED with optic neuropathy from orbital CT images.

Conclusions:

• The developed deep learning model accurately detects Thyroid Eye Disease and associated compressive optic neuropathy from orbital CT scans.
• The AI model significantly outperformed human expert grading in diagnostic accuracy.
• This AI tool has the potential to assist healthcare providers in early TED detection and patient stratification for specialist referral.