Related Experiment Video
Updated: Mar 29, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
2.4K
Hyper-Thyro Vision: An Integrated Framework for Hyperthyroidism Diagnostic Facial Image Analysis Based on Deep
Poonyisa Thepmangkorn1, Suchada Sitjongsataporn2
1The Electrical Engineering Graduate Program, Faculty of Engineering and Technology, Mahanakorn University of Technology, Nongchok, Bangkok 10530, Thailand.
Biomimetics (Basel, Switzerland)
|March 27, 2026
Summary
This study introduces an AI framework for detecting hyperthyroidism symptoms like exophthalmos and goiter using multi-modal image analysis. The system achieves high accuracy in identifying eye and neck abnormalities, improving diagnostic capabilities.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Ophthalmology
- Endocrinology
Background:
- Hyperthyroidism presents with characteristic ocular (exophthalmos) and neck (goiter) abnormalities.
- Accurate detection of these signs is crucial for timely diagnosis and management of hyperthyroidism.
- Existing diagnostic methods may lack the integrated, multi-modal approach for comprehensive assessment.
Purpose of the Study:
- To develop and evaluate an integrated, AI-driven, multi-modal framework for detecting hyperthyroidism-associated exophthalmos and goiter.
- To enhance diagnostic accuracy by jointly analyzing frontal facial and neck images.
- To mimic clinical visual assessment and physical examination through a dual-pathway deep learning architecture.
Main Methods:
- A dual-pathway deep learning framework processing frontal facial (eye regions) and neck images concurrently.
- Utilized YOLOv11s for eye region preprocessing, followed by a face mesh-based eye landmark (FMEL) approach and a sclera map unwrapping engine (SMUE) for quantitative scleral metrics.
- Employed a neck region of interest (ROI) prediction and neck μ-σ ensemble thresholding (NSET) algorithm for goiter assessment.
Main Results:
- The eye analysis achieved a mean average precision (mAP50) of 96.4% (98.6% for hyperthyroid class).
- SMUE demonstrated distinct morphological differences in scleral measurements between experimental and normal groups.
- The NSET algorithm achieved 92.0% mAP50 for swollen neck classification, outperforming baseline models with lower computational cost.
Conclusions:
- The proposed integrated multi-modal AI framework effectively detects hyperthyroidism-associated exophthalmos and goiter.
- The dual-pathway architecture and specific algorithms (SMUE, NSET) show high diagnostic performance.
- This approach offers a promising, computationally efficient tool for improving hyperthyroid abnormality detection.
