Towards Automated Eye Diagnosis: An Improved Retinal Vessel Segmentation Framework Using Ensemble Block Matching 3D...

Area of Science:

• Ophthalmology
• Medical Imaging
• Image Processing

Background:

• Accurate segmentation of retinal blood vessels is crucial for automated detection of vision-threatening eye diseases.
• Noisy and poorly illuminated fundus images present significant challenges for detecting fine vessels.
• Common noise types include additive and multiplicative (speckle) noise inherent to fundus imaging systems.

Purpose of the Study:

• To develop an efficient unsupervised strategy for retinal vessel segmentation in noisy fundus images.
• To improve the accuracy of eye disease classification by enhancing vessel detection.
• To address and mitigate artifacts introduced by noise reduction techniques.

Main Methods:

• Proposed an ensemble Block Matching 3D (BM3D) speckle filter for noise removal in fundus images.
• Developed a strategy to suppress checkerboard artifacts generated by the BM3D-speckle filter.
• Implemented an improved unsupervised segmentation approach utilizing the enhanced image data.

Main Results:

• Achieved high sensitivity rates: 82.88% (DRIVE), 81.41% (STARE), and 82.03% (CHASE_DB1).
• Boosted accuracy to 95.41% (DRIVE), 95.70% (STARE), and 95.61% (CHASE_DB1).
• Demonstrated superior performance on images with pathologies compared to state-of-the-art methods.

Conclusions:

• The proposed ensemble BM3D filtering and unsupervised segmentation effectively enhances fine vessel detection in noisy fundus images.
• The method offers improved accuracy and robustness for automated eye disease classification.
• This approach shows significant potential for clinical applications in ophthalmology.