Klasifikasi Multikelas Penyakit Mata dengan Modalitas Citra Fundus Menggunakan Convolutional Neural Network (CNN) dan Visualisasi Grad-CAM

Ramadhani, Adelia Putri (2025) Klasifikasi Multikelas Penyakit Mata dengan Modalitas Citra Fundus Menggunakan Convolutional Neural Network (CNN) dan Visualisasi Grad-CAM. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211062-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Penyakit mata seperti Diabetic Retinopathy (DR), Glaucoma, dan Cataract merupakan penyebab utama kebutaan di dunia yang berdampak signifikan terhadap kualitas hidup penderitanya. Deteksi dini sangat penting untuk mencegah kerusakan retina yang progresif. Namun, diagnosis masih sangat bergantung pada evaluasi manual citra fundus oleh tenaga medis ahli yang jumlahnya terbatas, terutama di daerah dengan keterbatasan akses kesehatan. Oleh karena itu, penelitian ini mengembangkan sistem klasifikasi otomatis berbasis deep learning menggunakan arsitektur EfficientNetB3 yang dilengkapi dengan visualisasi Grad-CAM guna meningkatkan interpretabilitas hasil prediksi. Dataset yang digunakan adalah ODIR (Ocular Disease Intelligent Recognition), dengan lima kelas target, yaitu Normal, DR, Glaucoma, Cataract, dan Other Diseases, yang telah disederhanakan menjadi hanya mencakup penyakit Age-Related Macular Degeneration (AMD). Citra fundus diproses melalui tahapan pre-processing seperti Circular Border Cropping, RGB Channel Extraction, Image Resizing ke ukuran 300×300 piksel, CLAHE (Contrast Limited Adaptive Histogram Equalization), Median Filtering, serta augmentasi data. Model EfficientNetB3 diinisialisasi dengan bobot awal dari ImageNet dan dilatih ulang dengan menggunakan teknik fine-tuning pada 100 layer terakhir menggunakan learning rate sebesar 1e-6 dan dropout 0.3 untuk meningkatkan adaptasi terhadap karakteristik domain citra retina. Hasil pengujian menunjukkan bahwa model mencapai akurasi training sebesar 97.08%, akurasi validation sebesar 87.20%, dan akurasi testing sebesar 89.26%. Evaluasi metrik lainnya mencakup precision sebesar 89%, recall sebesar 89%, specificity sebesar 98%, F1-score sebesar 89%, kappa score sebesar 0.86, dan AUC sebesar 0.98. Visualisasi Grad-CAM menunjukkan aktivasi pada area retina yang relevan secara medis seperti makula dan disk optik, sehingga mendukung validitas hasil klasifikasi dan dapat membantu tenaga medis dalam proses diagnosis.
====================================================================================================================================
Eye diseases such as Diabetic Retinopathy (DR), Glaucoma, and Cataract are major causes of blindness worldwide and significantly affect the quality of life of patients. Early detection is essential to prevent progressive retinal damage. However, diagnosis still heavily relies on manual evaluation of fundus images by medical professionals, whose availability is limited, especially in areas with limited access to healthcare. Therefore, this study develops an automatic classification system based on deep learning using the EfficientNetB3 architecture, equipped with Grad-CAM visualization to enhance the interpretability of prediction results. The dataset used is ODIR (Ocular Disease Intelligent Recognition), with five target classes: Normal, DR, Glaucoma, Cataract, and Other Diseases, which has been simplified to include only Age-Related Macular Degeneration (AMD). Fundus images are processed through several pre-processing steps such as Circular Border Cropping, RGB Channel Extraction, Image Resizing to 300×300 pixels, CLAHE (Contrast Limited Adaptive Histogram Equalization), Median Filtering, and data augmentation. The EfficientNetB3 model is initialized with pretrained ImageNet weights and retrained using fine-tuning on the last 100 layers, with a learning rate of 1e-6 and dropout of 0.3 to better adapt to the specific characteristics of retinal images. The test results show that the model achieved a training accuracy of 97.08%, validation accuracy of 87.20%, and testing accuracy of 89.26%. Additional evaluation metrics include a precision of 89%, recall of 89%, specificity of 98%, F1-score of 89%, kappa score of 0.86, and AUC of 0.98. Grad-CAM visualization highlights activation in medically relevant retinal areas such as the macula and optic disc, supporting the validity of the classification results and potentially assisting medical professionals in the diagnostic process.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN, Fundus Image, Grad-CAM, Klasifikasi Multi-kelas, Penyakit Mata, Eye Disease, Multi-class Classification.
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine R Medicine > RE Ophthalmology R Medicine > RE Ophthalmology > RE48 Eye--Diseases. Ophthalmoscopy. T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Electrical Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Adelia Putri Ramadhani
Date Deposited:	04 Aug 2025 11:32
Last Modified:	04 Aug 2025 11:32
URI:	http://repository.its.ac.id/id/eprint/126211

Actions (login required)

View Item