Performance analysis of retinal features for diabetic retinopathy characterisation and diagnosis

2021 
Automated retinal feature extraction serves as a significant step for accurate diabetic retinopathy (DR) diagnosis and screening. Ophthalmologists highly rely on retinal fundus segmentations for the treatment of severe eye related abnormalities; glaucoma, strokes, occlusions and DR. DR caused due to prolonged retinal blood vessel deterioration and if it is left untreated, it may lead to severe vision complications. This work proposes a mathematical morphology based optical disc localisation and blood vessel extraction approach to determine vascular alterations and expedite accurate identification of pathological symptoms. Shape and pixel based features are evaluated for optical disc localisation and blood vessels obtained along with a new set of fractal features to determine the self similarity between the extracted blood vessels and ground truth vessels labelled by the professional experts. A three-fold analysis is done on two benchmark databases; DRIVE and STARE, consisting of visual, statistical and performance analysis of the proposed blood vessel extraction approach. The performance evaluation of the proposed approach yields average accuracies of 95.50% and 94.80% respectively, when tested on benchmark DRIVE and STARE databases. The accuracy values obtained are appreciable and comparable to the various researches done in this field.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    0
    References
    0
    Citations
    NaN
    KQI
    []