Klasifikasi Katarak Dengan Modalitas Citra Fundus Mengunakan Convolution Neural Network (CNN)

Katarina, Lisa (2026) Klasifikasi Katarak Dengan Modalitas Citra Fundus Mengunakan Convolution Neural Network (CNN). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221057_Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Katarak merupakan salah satu penyebab utama gangguan penglihatan pada populasi lansia yang dapat berkembang menjadi kebutaan permanen apabila tidak dideteksi dan ditangani sejak dini. Tantangan utama dalam proses diagnostik saat ini adalah ketergantungan pada penilaian manual yang bersifat subjektif dan memerlukan keahlian klinis yang tinggi. Untuk mengatasi keterbatasan tersebut, penelitian ini mengembangkan sebuah sistem klasifikasi katarak berbasis Convolutional Neural Network (CNN) dengan Hybrid Resnet34 memanfaatkan citra fundus mata. Metode yang digunakan meliputi ekstraksi fitur visual citra fundus yang mencakup perubahan struktur retina dan tingkat kekeruhan lensa. Selain itu, hasil segmentasi anatomi Optic Disc (OD) dan Optic Cup (OC), serta metadata untuk meningkatkan transparansi proses prediksi, penelitian ini menerapkan algoritma Gradient-weighted Class Activation Mapping (Grad-CAM) untuk memberikan visualisasi area citra yang menjadi fokus CNN. Teknik ini memvalidasi bahwa model benar-benar memperhatikan struktur mata yang relevan saat menghasilkan keputusan klasifikasi. Berdasarkan hasil pengujian komprehensif, sistem yang dikembangkan sudah stabil dan akurat. Model mencatatkan tingkat akurasi pelatihan sebesar 92,77%, validasi 95,11%, testing yang berada di angka 92,58%. Hasil ini dibuktikan lebih lanjut melalui evaluasi metrik macro pada data uji, yang menghasilkan presisi 92,55%, recall 92,58%, F1-Score 92,56%, serta spesifisitas yang sangat tinggi mencapai 97,53%. Nilai dari Area Under the ROC Curve (ROC AUC) sebesar 98,79%. Evaluasi ketepatan representasi klasifikasi menggunakan metrik Intersection over Union (IoU) juga menunjukkan hasil yang memuaskan, yakni 86,60% pada data pengujian dan 90,24% pada data validasi. Selisih metrik yang kecil antara fase pelatihan dan validasi (Gap Loss 0,0206) mengonfirmasi bahwa model beroperasi dengan stabilitas tinggi tanpa adanya indikasi overfitting.
=====================================================================================================================================
Cataracts are one of the leading causes of visual impairment in the elderly population, which can progress to permanent blindness if not detected and treated early. The primary challenge in the current diagnostic process is the reliance on manual assessment, which is highly subjective and requires extensive clinical expertise. To overcome these limitations, this study develops a cataract classification system based on a Convolutional Neural Network (CNN) Hybrid Resnet34 utilizing eye fundus images. The proposed method includes the extraction of visual features from fundus images, encompassing changes in retinal structure and the degree of lens opacity. Furthermore, the anatomical segmentation results of the Optic Disc (OD) and Optic Cup (OC), along with metadata, are integrated into the system. To enhance the transparency of the prediction process, this study applies the Gradient-weighted Class Activation Mapping (Grad-CAM) algorithm to provide visualizations of the image areas prioritized by the CNN. This technique validates that the model strictly focuses on the relevant anatomical structures of the eye when generating classification decisions. Based on comprehensive testing, the developed system has proven to be highly stable and accurate. The model recorded a training accuracy of 92.77%, validation accuracy of 95.11%, and a robust testing accuracy of 92.58%. This diagnostic reliability is further evidenced by macro-metric evaluations on the test data, yielding a precision of 92.55%, recall of 92.58%, F1-Score of 92.56%, and a remarkably high specificity of 97.53%. The Area Under the ROC Curve (ROC AUC) reached 98.79%. The evaluation of classification localization using the Intersection over Union (IoU) metric also demonstrated satisfactory results, achieving 86.60% on the testing data and 90.24% on the validation data. The minimal metric disparity between the training and validation phases (Gap Loss of 0.0206) confirms that the model operates with high stability without any indications of overfitting. Overall, this research successfully developed a classification system capable of effectively supporting the early screening process for cataracts.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kata kunci: : Convolutional Neural Network, Grad-CAM, Katarak, Klasifikasi citra medis, Optic disc, Optic Cup ; Keywords: Cataract, Convolutional Neural Network, Fundus imaging, Grad-CAM, Medical image classification, Optic cup (OC), Optic disc (OD).
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Lisa Katarina
Date Deposited:	31 Jul 2026 06:40
Last Modified:	31 Jul 2026 06:40
URI:	http://repository.its.ac.id/id/eprint/140598

Actions (login required)

View Item