RAG-FW: A hybrid convolutional framework for the automated extraction of
retinal lesions and lesion-influenced grading of human retinal pathology
Abstract
[Submitted in IEEE J-BHI]
Retinopathy refers to any
damage in the retina that causes visual impairments or even blindness.
Identification of retinal lesions plays a vital role in accurately
grading retinopathy and for its effective treatment. Optical coherence
tomography (OCT) imaging is the most popular non-invasive technique used
for the retinal examination due to its ability to screen abnormalities
in early stages. Many researchers have presented studies on OCT based
retinal image analysis over the past. However, to our best knowledge,
there is no framework yet available which can extract retinal lesions
from multi-vendor OCT scans and utilize them for the intuitive grading
of the human retina. To cater this lack, we propose a deep retinal
analysis and grading framework (RAG-FW). RAG-FW is a hybrid
convolutional framework which extracts retinal lesions such as
intra-retinal fluid, sub-retinal fluid, hard exudates, drusen and
chorioretinal abnormalities (including fibrotic scars and choroidal
neovascular membranes) from multi-vendor OCT scans. Furthermore, it
utilizes them for the lesion-influenced grading of retinopathy as per
the clinical standards. RAG-FW has been trained using 113,261 retinal
OCT scans from which 112,261 scans were used for training and 1,000
scans were used for the validation purposes. Furthermore, it has been
rigorously tested on 43,613 scans from five highly complex publicly
available datasets where it achieved the mean intersection-over-union
score of 0.8055 for extracting the retinal lesions and the
F1 score of 99.52% for correctly classifying the
retinopathy cases. The source code of RAG-FW is available at
http://biomisa.org/index.php/downloads/.