研究目的
To detect and classify diabetic retinopathy in lesions using multi-scale Local Binary Pattern (LBP) feature extraction and Support Vector Machine (SVM) classification.
研究成果
The proposed system provides an effective approach for the early detection of diabetic retinopathy in lesions using multi-scale LBP features and SVM classification, achieving high sensitivity, specificity, and accuracy. The detection of distinct variants such as exudates, hemorrhages, micro aneurysms, and neovascularization aids in diagnosing the disease in its early stages.
研究不足
The study does not address the problem caused during laser treatment and the manual setting of threshold values for each set of images in existing techniques.
1：Experimental Design and Method Selection:
The study employs a preprocessing method to detect the Optic Nerve Head (ONH) in the lesion, followed by feature extraction using multi-scale Local Binary Pattern (LBP) algorithm and Gabor convolution. SVM classification is then used to locate hemorrhages and exudates, with a probabilistic multi-label lesion classification performed to categorize the results.
2：Sample Selection and Data Sources:
The input image database is taken from the publicly available MESSIDOR project, consisting of 1200 eye fundus color numerical images.
3：List of Experimental Equipment and Materials:
A color video 3CCD camera on a Topcon TRC NW6 non-mydriatic retinograph with a 45 degree field of view was used for image acquisition.
4：Experimental Procedures and Operational Workflow:
The process includes ROI extraction, preprocessing, feature extraction using multi-scale LBP, SVM classification, and probabilistic multi-label lesion classification.
5：Data Analysis Methods:
Performance is calculated based on sensitivity, accuracy, and specificity. The study also employs five metrics for multi-label classification analysis.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
