Multi-level deep supervised networks for retinal vessel segmentation.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal blood vessel changes indicate serious conditions like diabetes and hypertension.
• Segmenting these vessels in retinal images is difficult due to low contrast and complex structures.

Purpose of the Study:

• To develop a neural network-based method for accurate retinal blood vessel segmentation.
• To overcome challenges like low contrast, intricate topology, and pathologies in retinal images.

Main Methods:

• A deep supervised fully convolutional network utilizing multi-level hierarchical features.
• Incorporation of deep supervision with auxiliary classifiers to enhance feature discrimination and prevent gradient vanishing.
• Leveraging transfer learning to address limited medical training data and avoid hand-crafted features or preprocessing.

Main Results:

• The method was evaluated on DRIVE, STARE, and CHASE_DB1 databases.
• Achieved superior or comparable performance to state-of-the-art methods with significantly faster processing speeds.
• Cross-training experiments confirmed the robustness of the approach.

Conclusions:

• The proposed method accurately segments retinal vessels at high speed.
• It is easily adaptable for other biomedical segmentation tasks.