研究目的
To classify the input images accurately into normal, moderate or severe categories, correctly predict and detect the presence of DME, and compute the accuracy of detection and compare with the earlier methods described in the literature.
研究成果
The proposed method for classification of severity of DME in retinal color fundus images resulted in an accuracy of 95.2% using ETDRS rule and 98% using a random forest classifier. The method incorporates the extraction of various textural features and considers the count and the size of exudates, increasing the accuracy of DME detection.
研究不足
The efficiency and accuracy could be improved by using hybrid noise removal techniques for better image analysis and better segmentation methods for detecting the OD, macula, and the HE.
1：Experimental Design and Method Selection:
The study uses morphological operations for OD and macula detection, and hard exudates segmentation. Exudates are classified using early treatment diabetic retinopathy standard (ETDRS) and a random forest classifier.
2：Sample Selection and Data Sources:
Data is taken from the MESSIDOR dataset, consisting of 1200 retinal fundus color images, with 700 images considered for the study.
3：List of Experimental Equipment and Materials:
MATLAB for functionality development and WEKA for simulation.
4：Experimental Procedures and Operational Workflow:
The process includes data gathering, pre-processing, ROI detection, HE segmentation, feature extraction, and classification.
5：Data Analysis Methods:
Performance is evaluated with metrics like accuracy, sensitivity, specificity, and accuracy.
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
