A novel retinal vessel detection approach based on multiple deep convolution neural networks.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Retinal vessel (RV) detection is critical for diagnosing eye diseases using fundus images.
• Challenges in RV detection include variations in vessel morphology and low image quality.

Purpose of the Study:

• To develop an efficient and accurate method for retinal vessel detection in fundus images.
• To address the limitations of existing methods in handling noisy and low-contrast images.

Main Methods:

• A Multiple Deep Convolutional Neural Network (MDCNN) framework was developed for RV detection.
• The MDCNN was trained using an incremental learning strategy on a limited dataset.
• A voting procedure was employed to obtain final classification results.

Main Results:

• The MDCNN achieved high accuracy (95.97-96.13%) and AUC scores (0.9726-0.9737) on the DRIVE dataset.
• On the STARE dataset, the MDCNN obtained 95.39% accuracy and 0.9539 AUC score.
• The proposed method demonstrated competitive performance compared to state-of-the-art techniques.

Conclusions:

• The MDCNN offers superior performance for retinal vessel segmentation compared to existing methods.
• The proposed approach eliminates the need for a preprocessing stage, directly utilizing color fundus images.