Severity Classification of Non-Proliferative Diabetic Retinopathy Using Convolutional Support Vector Machine

Putra, Ricky Eka and Tjandrasa, Handayani and Suciati, Nanik Severity Classification of Non-Proliferative Diabetic Retinopathy Using Convolutional Support Vector Machine. International Journal of Intelligent Engineering and Systems.

Text Jurnal_05111160012003_Ricky Eka Putra.pdf - Other Restricted to Repository staff only Download (811kB)

Abstract

Diabetic retinopathy merupakan salah satu penyakit komplikasi dari
diabetes mellitus yang menyerang mata manusia. Tingkat resiko terbesar dari
penyakit ini adalah kebutaan. Non-Proliferative Diabetic Retinopathy merupakan
salah satu jenis diabetic retinopathy pada tahap awal dengan ditandai tidak adanya
pertumbuhan pembuluh darah baru. Tingkat keparahan Non-Proliferative Diabetic
Retinopathy dapat dibedakan menjadi tiga yakni mild, moderate, dan severe.
Tingkat keparahan dari penderita diabetic retinopathy perlu diketahui lebih dini
agar dapat dilakukan penanganan secepatnya.
Pengelompokan Non-Proliferative Diabetic Retinopathy melibatkan dua
cara yakni secara konvensional dan inkonvesional. Pada proses pendeteksian
memerlukan segmentasi lesi-lesi tersebut, antara lain microaneurysms,
hemorrhage, dan exudates. Proses segmentasi tersebut menggunakan metode
morfologi matematika yang merupakan metode konvensional. Selanjutnya, proses
klasifikasi tingkat keparahan diabetic retinopathy menggunakan metode Adaptive
Neuro Fuzzy Inference System dengan berdasarkan fitur exudates.
Selain itu, klasifikasi tingkat keparahan Non-Proliferative Diabetic
Retinopathy menggunakan cara inkonvensional dapat dilakukan dengan
memanfaatkan deep learning. Pra-proses dalam klasifikasi ini memanfaatkan tiga
metode Adaptive Histogram Equalization, Morphology Contrast Enhancement,
dan Homomorphic untuk meningkatkan kontras citra. Kemudian metode
Convolutional Neural Network dengan arsitektur GoogLeNet, ResNet18,
ResNet50, dan ResNet101 digunakan sebagai pengekstraksian fitur serta
melakukan pereduksian fitur menggunakan Relief dan Principal Component
Analysis. Selanjutnya, metode Support Vector Machine Naïve Bayes digunakan
sebagai classifier pada proses klasifikasi ini.
Hasil akurasi terbaik klasifikasi yang diperoleh dengan menggunakan
cara konvensional adalah 85.06%, sedangkan hasil akurasi terbaik pada cara
inkonvensional diperoleh dengan menggunakan metode pra-proses Homomorphic
dan Morphology Contrast Enhancement serta relief sebagai reduksi fiturnya. Nilai
akurasi terbaik yang dapat dicapai dengan menerapkan metode-metode tersebut
adalah 87.5%
=============================================================================================
Diabetic retinopathy is a complication of diabetes mellitus that attacks
the eyes of human. The greatest risk level from this disease is blindness. NonProliferative
Diabetic
Retinopathy
is
one
type
of
diabetic
retinopathy
in
the
early

stages

with the absence of new blood vessel growth. The severity of NonProliferative

Diabetic Retinopathy can be divided into three namely mild,
moderate, and severe. The severity of diabetic retinopathy patients needs to be
known early to get some treatments as so as possible.
Non-Proliferative Diabetic Retinopathy grouping is done in two ways
namely conventional and unconventional methods. In the first way, this detection
process requires segmentation of lesions, including microaneurysms, hemorrhage,
and exudates. The segmentation process uses mathematical morphology which is
a conventional method. Furthermore, the severity classification process of diabetic
retinopathy implements the Adaptive Neuro Fuzzy Inference System based on the
features of exudates.
In addition, the classification of the severity of Non-Proliferative
Diabetic Retinopathy using unconventional methods can be done by utilizing deep
learning method. Pre-processing in this classification utilizes three methods of
Adaptive Histogram Equalization, Morphology Contrast Enhancement, and
Homomorphic to improve the image contrast. Then, the Convolutional Neural
Networks with GoogLeNet, ResNet18, ResNet50, and ResNet101 architectures
are used for feature extraction and feature reduction are performed using Relief
and Principal Component Analysis. The Support Vector Machine-Naïve Bayes
method is used as a classifier in this classification process.
The best result of this study obtained by using conventional methods is
85.06%, while the best result in unconventional methods is obtained by using
Homomorphic and Morphological Contrast Enhancement methods and Relief as a
feature reduction. The best accuracy that can be achieved by applying these
methods is 87.5%.

Item Type:	Article
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science. EDP Q Science > QA Mathematics > QA76 Computer software Q Science > QA Mathematics > QA76.6 Computer programming. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Depositing User:	Ricky Eka Putra
Date Deposited:	25 Aug 2020 05:24
Last Modified:	29 Nov 2023 08:02
URI:	http://repository.its.ac.id/id/eprint/80816

Actions (login required)

View Item