研究目的
To propose a low-cost automated system for diagnosing glaucoma and diabetic retinopathy using digital eye fundus images, feature extraction, and machine learning classifiers to enable mass screening and reduce diagnosis costs.
研究成果
The proposed automated diagnosis system effectively classifies eye fundus images into healthy, glaucoma, and diabetic retinopathy categories with high accuracy, sensitivity, and specificity using ANN and SVM classifiers. It is cost-effective and integrable with medical informatics systems, making it suitable for mass screening. Further enhancements with more data and features could improve performance.
研究不足
The study relies on publicly available databases, which may not cover all variations of disease progression. The system's performance could be affected by image quality and database diversity. Future improvements could include more features, larger databases, and advanced machine learning algorithms to enhance accuracy and robustness.
1：Experimental Design and Method Selection:
The study uses a combination of image processing techniques (e.g., histogram equalization, median filter, top-hat filter, Otsu's thresholding) for ROI extraction and feature extraction (LBP, Gabor filter, statistical features, color features). Principal Component Analysis (PCA) is employed for feature selection, and classifiers (Artificial Neural Network and Support Vector Machine) are used for classification into healthy, glaucoma, and diabetic retinopathy classes.
2：Sample Selection and Data Sources:
Publicly available digital retinal image databases are used, including High-Resolution Fundus (HRF) Dataset, DIARETDB0 Dataset, ORIGA Dataset, STARE Dataset, and DRISHTI-GS Dataset. These datasets provide images for healthy, glaucoma, and diabetic retinopathy cases.
3：List of Experimental Equipment and Materials:
A computer with a 2.3 GHz Core i5-2410M Processor and 12 GB of RAM is used for processing. MATLAB? 2018 is the software for implementing ANN and SVM classifiers. No specific hardware for image capture is detailed, as databases are pre-existing.
4：3 GHz Core i5-2410M Processor and 12 GB of RAM is used for processing. MATLAB? 2018 is the software for implementing ANN and SVM classifiers. No specific hardware for image capture is detailed, as databases are pre-existing.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The process involves pre-processing (green channel extraction, histogram equalization, median filtering, top-hat filtering, thresholding, morphological closing), feature extraction (LBP, statistical features of LBP, color features, Gabor filter features), PCA for feature selection, and classification using ANN and SVM with different feature sets (5, 10, 15, 31 features). Performance is evaluated using accuracy, sensitivity, specificity, and time consumption metrics.
5：Data Analysis Methods:
Statistical indices (true positive, false positive, false negative, true negative) are calculated to compute accuracy, sensitivity, and specificity. Confusion matrices and ROC curves are generated for performance analysis. Time consumption is measured for each classifier and feature set.
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
