研究目的
To analyze variations in retinal vasculature for predicting the risk of stroke using retinal fundus images and a support vector machine classifier.
研究成果
The research demonstrates that retinal vasculature analysis can effectively predict stroke risk, with SVM achieving 93% accuracy using a linear kernel. This non-invasive approach has potential for early stroke prognosis, but further validation with larger datasets and advanced computational tools is needed.
研究不足
The study uses a limited dataset (130 images total), manual detection of the optic disc center, and relies on specific software tools (MATLAB, Vampire). Larger databases and more automated methods could improve generalizability and accuracy.
1：Experimental Design and Method Selection:
The study uses retinal fundus images to compute features such as fractal dimension, branching angle, branching coefficient, asymmetry factor, and optimality ratio for arteries and veins. These features are input into a support vector machine (SVM) classifier for stroke prediction. The methodology includes image preprocessing, binarization, skeletonization, and feature extraction using the Vampire annotation tool.
2：Sample Selection and Data Sources:
Retinal images were collected from Sree Gokulam Medical Centre and Research Foundation, Kerala, with 80 control (healthy) subjects and 50 stroke subjects.
3：List of Experimental Equipment and Materials:
Retinal fundus imaging equipment (not specified), Vampire annotation tool for image analysis, MATLAB software for implementation.
4：Experimental Procedures and Operational Workflow:
Images are converted to grayscale, preprocessed with adaptive histogram equalization, binarized, and skeletonized. Features are computed in specific zones around the optic disc (Zone A, B, C). The SVM classifier is trained and tested with 50% of the data for each group.
5：Data Analysis Methods:
Statistical analysis of feature means and standard deviations, SVM classification with linear, quadratic, and polynomial kernels to determine 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
