研究目的
Investigating the therapeutic effects of a specific herbal medicine on a particular disease.
研究成果
The proposed methodology for precise detection of bright lesions on retinal images using a CNN showed promising results, despite the limitations in the dataset size. The study suggests further research with parallel combinations of more CNNs, each trained for one type of DR symptom separately, and an analysis of training data to find the most efficient prevalence of different types of data.
研究不足
The study was limited by the small number of precisely marked images available for testing (only two). Additionally, the accuracy of 75.83% may not be sufficient from a statistical point of view, and the study suggests that a larger dataset of marked and healthy images would be necessary for more significant results.
1：Experimental Design and Method Selection:
The study uses a convolutional neural network (CNN) for the classification of bright lesions (soft and hard exudates) on retinal images. The methodology includes the design of a CNN with four convolutional layers for feature extraction and two fully connected layers for final classification.
2：Sample Selection and Data Sources:
The study uses retinal images from the Messidor database, which were divided into smaller blocks of 128x128 pixels. 29 images precisely marked by an ophthalmologist were used, with 27 for training and 2 for testing.
3：List of Experimental Equipment and Materials:
The study does not specify the equipment and materials used beyond the Messidor database and the CNN.
4：Experimental Procedures and Operational Workflow:
The input retinal images were divided into blocks, and each block was classified by the CNN four times (original and three mirror images) to increase reliability. Positive classifications were summed and weighted.
5：Data Analysis Methods:
The study evaluates the results using sensitivity, specificity, and accuracy measures.
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
