Related Experiment Video
Updated: Oct 29, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
Automatic detection of papilledema through fundus retinal images using deep learning
Tanzila Saba1, Shahzad Akbar2, Hoshang Kolivand3,4
1Artificial Intelligence & Data Analytics (AIDA) Lab CCIS, Prince Sultan University, Riyadh, 11586, Saudi Arabia.
Microscopy Research and Technique
|July 8, 2021
Summary
This study introduces an automated deep learning system for detecting and grading papilledema, a vision-threatening condition caused by increased intracranial pressure. The system achieves high accuracy in identifying papilledema from retinal images, aiding early diagnosis and treatment.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Papilledema, characterized by optic nerve swelling due to elevated intracranial pressure, can lead to vision loss.
- Retinal nerve fiber layer (RNFL) opacification is a key abnormality in papilledema, detectable via fundus imaging.
- Early detection and grading of papilledema are crucial for timely intervention and preventing blindness.
Purpose of the Study:
- To develop and evaluate a deep learning-based automated system for detecting and grading papilledema using fundus retinal images.
- To utilize U-Net and Dense-Net architectures for accurate classification and severity assessment of papilledema.
- To establish a novel, automated approach for clinical application in diagnosing papilledema.
Main Methods:
- A two-stage deep learning approach was employed, starting with Dense-Net for optic disc classification (papilledema vs. normal).
- Fundus images classified as papilledema underwent Gabor filter preprocessing.
- U-Net segmented the vascular network to calculate vessel discontinuity index (VDI) and VDIP for grading papilledema severity.
Main Results:
- The Dense-Net model achieved high classification performance: 98.63% sensitivity, 97.83% specificity, and 99.17% accuracy.
- The U-Net model demonstrated excellent grading performance: 99.82% sensitivity, 98.65% specificity, and 99.89% accuracy for mild and severe papilledema.
- The system was validated on the STARE dataset, comprising 60 papilledema and 40 normal images.
Conclusions:
- The proposed deep learning system offers a highly accurate and automated method for both detection and grading of papilledema.
- This automated approach, leveraging U-Net and Dense-Net, represents a significant advancement for clinical diagnosis of papilledema.
- The system's high sensitivity and specificity suggest its potential to improve patient outcomes by enabling earlier and more precise management of papilledema.
