研究目的
To propose a deep neural network-based classification and segmentation (CAS) model to extract blood vessels in color retinal images, addressing the challenge of segmenting retinal vessels, particularly the capillaries.
研究成果
The proposed CAS model is able to segment retinal vessels more accurately than seven existing segmentation algorithms, achieving the highest AUC and top three performance in terms of accuracy, specificity, and sensitivity on the DRIVE database.
研究不足
The proposed CAS model has a very high computation complexity, taking more than 32 hours to perform training. However, segmentation is relatively fast, taking less than 20 seconds per image.
1：Experimental Design and Method Selection:
The study adopts a "divide and conquer" strategy, using a deep neural network-based classification and segmentation (CAS) model. It involves preprocessing retinal images, classifying patches into wide-vessel, middle-vessel, and capillary patches, and segmenting them using three U-Nets.
2：Sample Selection and Data Sources:
The DRIVE database, consisting of 20 training and 20 testing fundus retinal color images from 400 diabetic subjects, was used.
3：List of Experimental Equipment and Materials:
Intel Xeon CPU, NVIDIA Titan Xp GPU, 128 GB Memory, 120 GB SSD, and Keras 1.1.0 were used for training.
4：0 were used for training.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: The process includes retinal image preprocessing, patch extraction and classification, training of segmentation networks, and reconstruction of segmented vessels.
5：Data Analysis Methods:
Performance was evaluated based on accuracy, specificity, sensitivity, and area under curve (AUC).
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
