Region of Interest-Based Differential Segmentation Method for Diabetic Retinopathy

Abstract

Diabetic Retinopathy (DR) is a vision-impairing disease caused due to continuous high blood
sugar levels. Early detection of this syndrome prevents permanent blindness through
classified medical opinions. DR classification is based on the severity of blood vessels,
leakage, etc. The clinical analysis exploits computer-inferred results through intelligent
processing. In such a process, the fundamental problem is the non-differentiated intensity in
the selected Region of Interest (ROI). Such intensity problem increases the errors in
segmentation due to region mismatch. In this article, a novel Discerning Region Selection for
Differential Analysis (DRS-DA) is proposed. The proposed method identifies the variations
in the selected region’s intensity through the differential transform. This process performs
indecisive prediction by organizing the pixel positions. The organized pixels are correlated
with the input test images for similarity verification. The un-matching pixels are replaced by
partial differential transform for intensity detection. In this verification and substitution
process, deep recurrent learning is induced for identifying the failing transforms. The process
is repeated until; all the pixels of all the regions are classified. This helps to identify the errors
in discerning regions, improving the segmentation accuracy. The other metrics such as
identification time, precision, and error are analyzed for verifying the proposed method’s
performance.