Related Experiment Video
Updated: Aug 28, 2025
07:44
In vivo Structural Assessments of Ocular Disease in Rodent Models using Optical Coherence Tomography
Published on: July 24, 2020
3.0K
DMs-MAFM+EfficientNet: a hybrid model for predicting dysthyroid optic neuropathy.
Cong Wu1, Shijun Li2, Xiao Liu2
1School of Computer Science, Hubei University of Technology, Nanli Street 28, Wuhan, 430068, China. oidipous@hbut.edu.cn.
Medical & Biological Engineering & Computing
|September 21, 2022
Summary
A new hybrid deep learning model accurately identifies dysthyroid optic neuropathy (DON) in patients with thyroid-associated ophthalmopathy (TAO) using CT scans. This AI tool significantly aids in diagnosing this severe orbital disease, improving patient outcomes.
Area of Science:
- Ophthalmology
- Radiology
- Artificial Intelligence in Medicine
Background:
- Thyroid-associated ophthalmopathy (TAO) is a prevalent autoimmune orbital disease.
- Dysthyroid optic neuropathy (DON), the most severe form of TAO, affects 4%-8% of patients.
- Current diagnostic methods for suspected DON lack sufficient sensitivity and specificity, necessitating improved identification tools.
Purpose of the Study:
- To develop and validate a hybrid deep learning model for accurate identification of suspected DON patients.
- To enhance the diagnostic capabilities for a condition with currently limited clinical evaluation accuracy.
- To leverage computed tomography (CT) imaging for AI-driven DON diagnosis.
Main Methods:
- A hybrid deep learning model integrating a novel double multiscale and multi attention fusion module (DMs-MAFM) with a deep convolutional neural network was proposed.
- The DMs-MAFM employs multiscale feature fusion and enhanced channel/spatial attention mechanisms to capture subtle image features.
- The model was trained and evaluated for its ability to identify suspected DON patients using CT data.
Main Results:
- The hybrid deep learning model achieved high diagnostic performance: Accuracy 96%, Specificity 99.5%, Sensitivity 94%, Precision 98.9%, and F1-score 96.4%.
- The DMs-MAFM effectively captured features of tiny objects crucial for DON identification.
- Expert evaluation confirmed the model's significant potential for clinical diagnosis and prediction of DON.
Conclusions:
- The proposed hybrid deep learning model offers an efficient and accurate method for identifying suspected DON patients.
- This AI-driven approach shows considerable promise in assisting clinicians with the diagnosis and prediction of dysthyroid optic neuropathy.
- The study highlights the potential of advanced deep learning techniques in improving the management of complex autoimmune orbital diseases like TAO.
