PERFORMANCE OPTIMIZATION OF GLAUCOMA DETECTION BASED ON QUANTIFICATION OF OPTIC DISC AND OPTIC CUP CHARACTERISTICS ON RETINAL FUNDUS IMAGES

• المؤلف الرئيسي: Hanifa Suwandoko, Fadia
• التنسيق: Final Project
• اللغة: Indonesia
• الوصول للمادة أونلاين: https://digilib.itb.ac.id/gdl/view/66311

Glaucoma is eye disease which is caused by increase of intraocular pressure. The pressure is damaging optic nerve head and could lead to partially or even entirely loss of eyesight if there is no appropriate treatment. In Indonesia, only 51.4% of glaucoma cases were only examined after they were at the severe stage of glaucoma namely when there is already significant damage to the eye or even when vision has been greatly reduced. Therefore, early glaucoma detection is crucial so that glaucoma can be treated as fast as possible. Currently, there are many developments in this field that have been carried out using various types of approaches. One of the approaches is by quantification of optic disc and cup’s such as cup to disc ratio. In this approach, the segmentation of optic disc and cup play an important role in glaucoma detection. In a previous study, Almustofa (2021) had developed an automation of glaucoma detection method based on the quantification of optical disc and cup characteristic. However, it has only been tested on DrishtiGS and Refuge Training Set. In this study, the author has developed an optimization method using the addition of Refuge Validation set and Test set. The optimized method successfully localizes optic disc with 100.00 ± 0.00% proportion within region of interest for Drishti-GS and 99.83 ± 3.95% for Refuge. Optic disc segmentation achieves fscore 0.979 ± 0.005 for Drishti-GS and fscore 0.942 ± 0.026 for Refuge dataset. Furthermore, optic cup segmentation achieves the fscore 0.948 ± 0.020 in Drishti-GS dataset and fscore 0.843 ± 0.068 in Refuge dataset. Glaucoma classification achieves 0.880 accuracy, 0.667 specificity, and 1.000 sensitivity in Drishti-GS, and 0.955 accuracy, 0.981 specificity, and 0.725 sensitivity in Refuge datasets.