Related Experiment Video
Updated: Mar 1, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Machine learning techniques for diabetic macular edema (DME) classification on SD-OCT images.
Khaled Alsaih1,2, Guillaume Lemaitre1, Mojdeh Rastgoo1
1LE2I, CNRS, Arts et Métiers, Université Bourgogne Franche-Comté, 12 rue de la Fonderie, Le Creusot, France.
Biomedical Engineering Online
|June 9, 2017
Summary
This study introduces an automated framework to detect diabetic macular edema (DME) using spectral domain optical coherence tomography (SD-OCT) scans. The system achieved 87.5% sensitivity and specificity, offering a promising tool for DME diagnosis.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer Vision
Background:
- Spectral domain optical coherence tomography (SD-OCT) is crucial for diagnosing diabetic macular edema (DME) in ophthalmology.
- SD-OCT provides detailed visualization of retinal morphology and layers, aiding in DME detection.
Purpose of the Study:
- To develop and evaluate an automatic classification framework for SD-OCT volumes to distinguish between normal and DME cases.
- To investigate the effectiveness of various feature extraction, representation, and classification methods for DME detection.
Main Methods:
- Utilized a dataset of 32 SD-OCT volumes (16 normal, 16 DME) from the Singapore Eye Research Institute.
- Employed a pipeline involving pre-processing, feature detection (Histogram of Oriented Gradients, Local Binary Patterns), feature representation (Principal Component Analysis, Bag of Words), and classification.
Main Results:
- Evaluated individual and combined features, representation approaches, and classifiers.
- Achieved the best performance using Local Binary Patterns (LBP) features represented by Principal Component Analysis (PCA) and classified with a linear Support Vector Machine (SVM).
Conclusions:
- The proposed automated framework demonstrates high accuracy in classifying SD-OCT volumes for DME detection.
- The combination of LBP features, PCA, and linear SVM achieved a sensitivity and specificity of 87.5%, indicating its potential for clinical application.
