A Deep Learning Framework for Predicting Teprotumumab Treatment Response in Thyroid Eye Disease

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Thyroid eye disease (TED) is a debilitating autoimmune condition affecting the orbits.
• Accurate quantification of TED severity and prediction of treatment response are crucial for effective management.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) framework for quantifying TED severity.
• To create a predictive model for forecasting individual patient responses to teprotumumab therapy.

Main Methods:

• A retrospective study utilized a DL classification model integrating CT-based orbital volumetric features and clinical data.
• A severity scoring model converted classification probabilities into a continuous metric.
• A regression-based model predicted posttreatment severity using pretreatment variables.

Main Results:

• The DL classification model achieved high accuracy (0.81) and AUC (0.86-0.88) for TED severity.
• Teprotumumab treatment led to a significant mean improvement of 0.194 severity points (P < 0.001).
• The prediction model demonstrated strong performance with R^2 of 0.82 and 89% concordance with clinician assessment.

Conclusions:

• The DL framework shows promise for objective TED severity quantification.
• This approach offers a proof-of-concept for data-driven tools to support individualized TED treatment planning.