PCA-Based Retinal Vessel Tortuosity Quantification

Abstract

Rashmi TURIOR †a) , Student Member, Danu ONKAEW †b) , and Bunyarit UYYANONVARA †c) , Nonmembers SUMMARYAutomatic vessel tortuosity measures are crucial for many applications related to retinal diseases such as those due to retinopathy of prematurity (ROP), hypertension, stroke, diabetes and cardiovascular diseases. An automatic evaluation and quantification of retinal vascular tortuosity would help in the early detection of such retinopathies and other systemic diseases. In this paper, we propose a novel tortuosity index based on principal component analysis. The index is compared with three existant indices using simulated curves and real retinal images to demonstrate that it is a valid indicator of tortuosity. The proposed index satisfies all the tortuosity properties such as invariance to translation, rotation and scaling and also the modulation properties. It is capable of differentiating the tortuosity of structures that visually appear to be different in tortuosity and shapes. The proposed index can automatically classify the image as tortuous or non tortuous. For an optimal set of training parameters, the prediction accuracy is as high as 82.94% and 86.6% on 45 retinal images at segment level and image level, respectively. The test results are verified against the judgement of two expert Ophthalmologists. The proposed index is marked by its inherent simplicity and computational attractiveness, and produces the expected estimate, irrespective of the segmentation approach. Examples and experimental results demonstrate the fitness and effectiveness of the proposed technique for both simulated curves and retinal images. key words: tortuosity, retinopathy of prematurity (ROP), principal component analysis

Quantifying retinal blood vessels' tortuosity — Review

“…1) Curvature at a single point a long a blood vessel: Given a blood vessel segment (𝑆), as a plane curve, where 𝑆 is represented by centre line points represented by 𝑆 = [(𝑥 1 , 𝑦 1 ), (𝑥 2 , 𝑦 2 ), .., (𝑥 𝑛−2 , 𝑦 𝑛−2 )), (𝑥 𝑛−1 , 𝑦 𝑛−1 ), (𝑥 𝑛 , 𝑦 𝑛 )], and given parametrically in the Cartesian coordinates as (𝑡) = (𝑥(𝑡), 𝑦(𝑡)), The curvature 𝐶 at point t, 𝐶(𝑡), can be estimated as follows: Crystal S. Y. Cheung etal (2011) [24] Private Dataset Significant reductions in all vascular measurements were observed compared to pre-treatment * 6 Rashmi Turior (2012) [25] Private Dataset Achieved a classification rate of 73% * 5…”

“…So it can be compared on vessels that have a different length. The advantage of this formula is that it does not depend on segmentation of the vessel tree [25], [34].…”

Automatic Evaluation of Plus Disease in Retinopathy of Prematurity

“…Vessel partitioning to the skeletonized image entails vessel segmentation. The process involved in doing this, is the improved branching and ending point detection technique [23]. This provides us with a set of disconnected sub vessel samples (Fig.…”

Automated Detection and Tortuosity Characterization of Retinal Vascular Networks

“…The combination of curvature and arc-chord ratio is used for the measurement of tortuosity in retinal vascular network in [40]. Different classifiers such as k-nearest neighbour (kNN) and Naïve Bayesian with K-means algorithm are utilized to investigate vascular network tortuosity characterization in fundus images [41]. When compared with other techniques investigated in the study, KNN achieved the best mean accuracy value of 87.3%.…”