Blood Vessel Segmentation of Retinal Image Based on Dense-U-Net Network

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Accurate segmentation of retinal blood vessels is crucial for diagnosing fundus diseases.
• Traditional methods often suffer from segmentation errors and low accuracy.

Purpose of the Study:

• To develop an improved retinal blood vessel segmentation algorithm.
• To enhance accuracy and address limitations of existing methods using a combined deep learning approach.

Main Methods:

• Proposed a Dense-U-net model combining U-Net and Dense-Net architectures.
• Employed image preprocessing techniques including adaptive gamma correction and multi-scale morphological transformation.
• Utilized stochastic gradient descent to optimize the Dice loss function for improved segmentation.

Main Results:

• Achieved high performance metrics: specificity (0.9896), accuracy (0.9698), sensitivity (0.7931), and AUC (0.9738).
• Demonstrated significant improvement in segmenting both large and small retinal blood vessels.
• The method showed enhanced vascular feature information and artifact correction.

Conclusions:

• The proposed Dense-U-net model offers a robust and accurate solution for retinal blood vessel segmentation.
• This advancement has the potential to improve early detection and diagnosis of fundus-related eye conditions.
• The method's effectiveness in segmenting small vessels is particularly noteworthy for clinical applications.