Automatic optic disk detection in retinal images using hybrid vessel phase portrait analysis

Area of Science:

• Ophthalmology and Medical Imaging
• Biomedical Engineering
• Computer Vision

Background:

• Accurate detection and segmentation of the optic disk (OD) are crucial for diagnosing various eye conditions.
• Existing methods often struggle with image quality variations, particularly in retinal fundus images.
• Retinal blood vessel patterns offer valuable information for robust OD localization and segmentation.

Purpose of the Study:

• To develop and evaluate novel algorithms for automatic optic disk (OD) detection and segmentation.
• To improve OD detection accuracy, especially in low-quality retinal images.
• To assess the clinical utility of the developed methods for detecting diabetic retinopathy (DR).

Main Methods:

• Vessel vector-based phase portrait analysis (VVPPA) utilizing retinal blood vessel direction.
• A hybrid method (HM) combining VVPPA with the vessel transform (VT) for enhanced OD localization.
• Integration of scale space (SS) with VVPPA and HM for OD contour identification (SSVVPPAC, SSHMC).

Main Results:

• The hybrid method (HM) achieved 98% accuracy in optic disk localization, outperforming benchmark methods, particularly on poor-quality images.
• The SSHMC segmentation method demonstrated superior performance, achieving high positive predictive value (PPV) and sensitivity for poor-quality images.
• In a clinical setting, the HM demonstrated high efficacy in detecting diabetic retinopathy (DR) abnormalities, with 98.13% true positive rate.

Conclusions:

• The proposed VVPPA and HM algorithms offer a robust and accurate approach to optic disk detection and segmentation.
• These methods are particularly effective in challenging low-quality retinal images, enhancing diagnostic capabilities.
• The HM shows significant potential for clinical applications, including the automated detection of diabetic retinopathy using smartphone-based imaging.