TORTUOSITY MEASUREMENT OF RETINAL VESSELS AND CORNEAL NERVES

• Main Author: Chuang, Winnie
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70042
• Institution: Institut Teknologi Bandung

Changes in tortuosity degree of retinal vessels and corneal nerves may indicate the progression of various diseases, such as hypertension or diabetic neuropathy. Therefore, observing the changes in the tortuosity levels in retinal fundus image and IVCM (in vivo confocal microscopy) can be a good biomarker in diagnosing diseases. However, tortuosity assessments performed by ophthalmologists are usually subjective and time-consuming. Thus, there is a need for a universally accepted tortuosity measurement standard as well as an automatic tortuosity measurement method. Several studies have proposed parameters or methods for measuring the tortuosity of blood vessels and nerves of the cornea. However, these methods still have some drawbacks. Several tortuosity measurement parameters that were proposed in the previous literature still measure tortuosity partially, which means that they only describe tortuosity in terms of angle changes or curvature. In addition, some parameters produce unequal correlation values for arteries and veins vessels. Meanwhile, the parameters used for corneal nerves tortuosity measurement in other literature have not fully described the characteristics of the nerves in the image, so the accuracy of the resulting tortuosity level classification is still not very accurate. Therefore, it is necessary to find the right variation of measurement parameters and aggregation methods that can produce the aggregation of tortuosity values that can accurately represent the tortuosity level of the image. This study aims to propose tortuosity measurement parameters or methods that can increase the correlation between tortuosity assessments from ophthalmologists and the results from automatic tortuosity measurements. The parameter proposed for retinal blood vessels tortuosity measurement is a combined parameter from previous works of literature, which are based on the comparison of arc length and chord length of the curve structure as well as the angle values at each of the critical points of the curve. To calculate the tortuosity of the corneal nerves, 14 tortuosity parameters will be used to measure the tortuosity of each nerve segment. Then, 9 aggregation methods will be used to combine the tortuosity values of the nerves in the image into 364 values that will become the tortuosity feature values at the image level. The feature values will go through a series of feature selection processes so that some of the features that are the most discriminatory against the tortuosity class distribution of the dataset are selected. The proposed method for retinal blood vessels achieved a Spearman rank correlation coefficient of 0.886 for arteries and 0.884 for veins. The proposed method for the corneal nerves dataset achieved weighted accuracy of 0.883 for the CORN-3 dataset and 0.911 for the CCM-B dataset.