Automatic Screening of Retinal Lesions for Grading Diabetic Retinopathy
Muhammad Sharif1
and Jamal Hussain Shah1,2
1Department of
Computer Science, COMSATS University Islamabad, Pakistan
2University of Science and Technology of
China, China
Abstract:
Diabetic Retinopathy (DR) is a diabetical
retinal syndrom. Large number of patients have been suffered from blindness due
to DR as compared to other diseases. Priliminary detection of DR is a critical quest
of medical image processing. Retinal Biomarkers are termed as Microaneurysms
(MAs), Haemorrhages (HMAs) and Exudates (EXs) that are helpful to grade
Non-Proliferative DR (NPDR) at different stages. This research work contributes
an automatic design for the retinal lesions screening to grade DR system. The
system is comprised of unique preprocessing determination of biomarkers and formulation
of profile set for classification. During preprocessing, Contrast Limited
Adaptive Histogram Equalization (CLAHE) is utilized and Independent Component
Analysis (ICA) is extended with Curve Fitting Technique (CFT) to eliminate
blood vessels and optic disc as well as to detect biomarkers from the digital
retinal image. Subsequent, NPDR lesions based distinct eleven features are deduced
for the purpose of classification. Experiments are performed using a fundus
image database. The proposed method is appropriate for initial grading of DR.
Keywords: DR, CLAHE, ICA, CFT, biomarkers.