Clinical Scorecard: Interpretable Machine Learning Approaches for Classifying Severity in Thyroid Eye Disease Through Orbital Anatomical Characteristics
At a Glance
Category
Detail
Condition
Thyroid Eye Disease (TED)
Key Mechanisms
MRI-derived anatomical measurements including ocular protrusion and extraocular muscle thickness.
Target Population
Patients with Thyroid Eye Disease
Care Setting
Ophthalmic imaging and assessment
Key Highlights
Random Forest with class weighting achieved the highest AUC of 0.811.
Feature importance analysis ranked ocular protrusion as the top predictor.
Controlling for longitudinal redundancy significantly impacts model evaluation.
The framework integrates measurable anatomical parameters for severity stratification.
Standardized quantification is emphasized for reproducibility in medical AI.
Guideline-Based Recommendations
Diagnosis
Use MRI to assess anatomical changes in TED.
Management
Integrate objective anatomical parameters with clinical assessment for TED severity.
Monitoring & Follow-up
Utilize longitudinal MRI scans to track disease progression.
Risks
Subjectivity in clinical scoring may lead to inter-observer variability.
Patient & Prescribing Data
443 patients with Thyroid Eye Disease analyzed.
Machine learning models can enhance the objectivity of TED severity assessment.
Clinical Best Practices
Employ interpretable machine learning frameworks for clinical decision support.
Standardize data structuring to improve model generalizability.
Incorporate anatomical metrics in routine TED assessments.
Susan Gromacki, OD, MS, FAAO, FSLS, Dipl AAO, and Clark Chang, OD, FAAO, FSLS, reviewed the latest developments in diagnosis, imaging, contact lenses, corneal cross-linking (CXL), and surgical management for KC.
