Abstract:
Hypertension-related Retinopathy (HR) is the medical term for an illness that can have
an effect on the retina. The severity and length of time that a person has had hyperten sion is strongly correlated with the prevalence of HR-eye disease. The following are the
ocular pathological abnormalities that are related with HR: constriction of the arterio lar blood vessels, macular edoema, retinal haemorrhage, cotton wool patches, and blood
vessels. In order to forestall the development of blindness, it is absolutely necessary to
detect and evaluate HR as quickly as humanly possible. The number of computer-aided
diagnostic systems (CADx) currently available on the market that are able to diagnose
HR is extremely limited. In addition to having in-depth expertise in the field, being
able to operate these CADx systems needed sophisticated knowledge of the various im age processing techniques. In recent years, it has been discovered that categorization
algorithms have an unsatisfactory level of accuracy.
To overcome these challenges, a brand new CAD-HR system that consists of a depthwise
separable convolutional neural network (DSC) with residual connection and a linear
support vector machine (LSVM) has been constructed. This system was developed
in order to meet the aforementioned challenges. In order to increase the total size of
the datasets, the data augmentation technique is first used to retina graphics. This
is done with the intention of improving the quality of the data. After that, the DSC
method is used on photographs of the retina in order to obtain consistent properties. In
the final step, an LSVM classifier is utilised to make the determination of whether or
not the retinal samples represent HR or non-HR. The statistical investigation of 9,500
renography images obtained from two sources that are available to the public and one
source that is kept private in order to evaluate the accuracy of the implementation of
the CAD-HR system utilising several performance metrics including accuracy, specificity,
and the AUC-ROC curve. These performance metrics were used in order to evaluate the proficiency of the implementation of the CAD-HR system. It has been established
that the DSC-HR model takes less computing time and fewer parameters in order to
accurately categorise HR phases.
