biaya tinggi pemeriksaan sering kali menjadi hambatan, terutama di Indonesia. Kamera fundus konvensional, meskipun efektif, memiliki harga yang mahal dan kurang portabel, membatasi aksesibilitas di daerah terpencil. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi katarak otomatis menggunakan kamera fundus berbasis smartphone, yang menawarkan solusi lebih ekonomis dan portabel dibandingkan perangkat konvensional. Penelitian ini mengevaluasi kinerja tiga algoritma jaringan syaraf tiruan yaitu, Backpropagation Neural Network (BPNN), Probabilistic Neural Network (PNN), dan Radial Basis Function Neural Network (RBFNN), dalam klasifikasi katarak. Metode penelitian meliputi pra-pengolahan citra, segmentasi optic disc, ekstraksi ciri tekstur menggunakan Gray-Level Co-occurrence Matrix (GLCM) dan Filter Gabor, serta klasifikasi tingkat keparahan katarak ke dalam empat kategori, yaitu retina normal, katarak ringan (mild), katarak sedang (medium), dan katarak berat (severe). Hasil pelatihan menunjukkan rerata nilai akurasi sistem sebesar 96,35%, dengan kinerja terbaik pada ekstraksi ciri GLCM menggunakan PNN (100%) dan Filter Gabor menggunakan PNN (96,88%). Sensitivitas tertinggi dalam pelatihan dicapai oleh metode GLCM dan PNN (100%) untuk kategori katarak normal dan berat. Pada pengujian, sistem mencapai nilai rerata akurasi sebesar 77,98%, dengan hasil terbaik pada ekstraksi ciri GLCM menggunakan PNN (89,29%). Sensitivitas tertinggi pada pengujian diperoleh dengan metode GLCM dan PNN (89,29%) untuk katarak ringan dan berat, sementara spesifisitas tertinggi dicapai oleh GLCM dan BPNN (95,24%) untuk katarak normal. Temuan ini menunjukkan bahwa sistem berbasis smartphone ini tidak hanya meningkatkan aksesibilitas diagnosis katarak di daerah terpencil tetapi juga memberikan akurasi yang kompetitif dengan solusi konvensional. Absctract Eye health, particularly cataract diagnosis, is a crucial aspect of individual well-being. However, the high cost of examinations often poses a barrier, especially in Indonesia. Conventional fundus cameras, while effective, are expensive and less portable, limiting accessibility in remote areas. This research aims to develop an automatic cataract classification system using smartphone-based fundus cameras, offering a more cost-effective and portable solution compared to conventional devices. The study evaluates the performance of three neural network algorithms: Backpropagation Neural Network (BPNN), Probabilistic Neural Network (PNN), and Radial Basis Function Neural Network (RBFNN) for cataract classification. The research methodology includes image preprocessing, optic disc segmentation, texture feature extraction using Gray-Level Co-occurrence Matrix (GLCM) and Gabor Filter, and classification of cataract severity into four categories: normal retina, mild cataract, medium cataract, and severe cataract. Training results show an average system accuracy of 96.35%, with the best performance on GLCM feature extraction using PNN (100%) and Gabor Filter using PNN (96.88%). The highest sensitivity in training was achieved by GLCM and PNN (100%) for normal and severe cataract categories. During testing, the system achieved an average accuracy of 77.98%, with the best results for GLCM feature extraction using PNN (89.29%). The highest sensitivity in testing was obtained with GLCM and PNN (89.29%) for mild and severe cataracts, while the highest specificity was achieved by GLCM and BPNN (95.24%) for normal cataracts. These findings indicate that the smartphone-based system not only enhances cataract diagnosis accessibility in remote areas but also provides competitive accuracy compared to conventional solutions.
English
biaya tinggi pemeriksaan sering kali menjadi hambatan, terutama di Indonesia. Kamera fundus konvensional, meskipun efektif, memiliki harga yang mahal dan kurang portabel, membatasi aksesibilitas di daerah terpencil. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi katarak otomatis menggunakan kamera fundus berbasis smartphone, yang menawarkan solusi lebih ekonomis dan portabel dibandingkan perangkat konvensional. Penelitian ini mengevaluasi kinerja tiga algoritma jaringan syaraf tiruan yaitu, Backpropagation Neural Network (BPNN), Probabilistic Neural Network (PNN), dan Radial Basis Function Neural Network (RBFNN), dalam klasifikasi katarak. Metode penelitian meliputi pra-pengolahan citra, segmentasi optic disc, ekstraksi ciri tekstur menggunakan Gray-Level Co-occurrence Matrix (GLCM) dan Filter Gabor, serta klasifikasi tingkat keparahan katarak ke dalam empat kategori, yaitu retina normal, katarak ringan (mild), katarak sedang (medium), dan katarak berat (severe). Hasil pelatihan menunjukkan rerata nilai akurasi sistem sebesar 96,35%, dengan kinerja terbaik pada ekstraksi ciri GLCM menggunakan PNN (100%) dan Filter Gabor menggunakan PNN (96,88%). Sensitivitas tertinggi dalam pelatihan dicapai oleh metode GLCM dan PNN (100%) untuk kategori katarak normal dan berat. Pada pengujian, sistem mencapai nilai rerata akurasi sebesar 77,98%, dengan hasil terbaik pada ekstraksi ciri GLCM menggunakan PNN (89,29%). Sensitivitas tertinggi pada pengujian diperoleh dengan metode GLCM dan PNN (89,29%) untuk katarak ringan dan berat, sementara spesifisitas tertinggi dicapai oleh GLCM dan BPNN (95,24%) untuk katarak normal. Temuan ini menunjukkan bahwa sistem berbasis smartphone ini tidak hanya meningkatkan aksesibilitas diagnosis katarak di daerah terpencil tetapi juga memberikan akurasi yang kompetitif dengan solusi konvensional. Absctract Eye health, particularly cataract diagnosis, is a crucial aspect of individual well-being. However, the high cost of examinations often poses a barrier, especially in Indonesia. Conventional fundus cameras, while effective, are expensive and less portable, limiting accessibility in remote areas. This research aims to develop an automatic cataract classification system using smartphone-based fundus cameras, offering a more cost-effective and portable solution compared to conventional devices. The study evaluates the performance of three neural network algorithms: Backpropagation Neural Network (BPNN), Probabilistic Neural Network (PNN), and Radial Basis Function Neural Network (RBFNN) for cataract classification. The research methodology includes image preprocessing, optic disc segmentation, texture feature extraction using Gray-Level Co-occurrence Matrix (GLCM) and Gabor Filter, and classification of cataract severity into four categories: normal retina, mild cataract, medium cataract, and severe cataract. Training results show an average system accuracy of 96.35%, with the best performance on GLCM feature extraction using PNN (100%) and Gabor Filter using PNN (96.88%). The highest sensitivity in training was achieved by GLCM and PNN (100%) for normal and severe cataract categories. During testing, the system achieved an average accuracy of 77.98%, with the best results for GLCM feature extraction using PNN (89.29%). The highest sensitivity in testing was obtained with GLCM and PNN (89.29%) for mild and severe cataracts, while the highest specificity was achieved by GLCM and BPNN (95.24%) for normal cataracts. These findings indicate that the smartphone-based system not only enhances cataract diagnosis accessibility in remote areas but also provides competitive accuracy compared to conventional solutions.
Affiliations come from OpenAlex; journal OAI-PMH metadata carries none. Authors without an OpenAlex ID are matched by name, which can merge or split people.