Digital image analysis in retinopathy of prematurity: A comparison of vessel selection methods

An experimental assessment of five indices of retinal vessel tortuosity with the RET-TORT public dataset

“…Our aim was to compare the performance of the five algorithms selected, using the Spearman rank correlation of automatic and clinical rankings, as done elsewhere [14], [18]. Values obtained by each index for arteries and veins were obtained separately.…”

“…Grading results have been shown to vary with ophthalmologist, e.g., [8]. An objective, repeatable and quantitative assessment of vessel tortuosity has been long deemed desirable, as it sometimes informs important treatment decisions (e.g., ROP-plus disease [18]), and for biomarker research [19]. Several quantitative tortuosity indices and associated automatic algorithms have been proposed [9], [10], [11], [12], [17] but, due to the lack of public, large annotated datasets, ideally disease-specific, it remains very difficult to compare algorithms comprehensively and fairly [13].…”

“…The indices were selected considering together their performance (as reported in the original papers), recency and impact on the field (citations, take-up). Following the established methodology [10], [14], [18], we estimate the tortuosity values using all indices for all vessels in both sets (artery and vein), rank them from least to most tortuous, and applied to retinal images [12], compute the Spearman correlation, ρ, with the ranking obtained from clinicians. In the following, we summarize the tortuosity indices (Section II) and the data set used (Section III), describe the experimental protocol (Section IV), report and discuss results (Section V) and offer conclusions (Section VI).…”

Computer-Based Image Analysis for Plus Disease Diagnosis in Retinopathy of Prematurity: Performance of the “i-ROP” System and Image Features Associated With Expert Diagnosis

“…15,[27][28][29][30] Additionally, the observation that the point-based features of acceleration and curvature outperformed segmentbased features may suggest that the i-ROP system may perform similarly without requiring manually segmented images as inputs. This may be due to the fact that GMM fitting can be more robust with pointbased features, since the number of vessel points is much larger than the number of vessel segments in an image.…”

Methods for digital analysis of human vascular system. Literature review

“…Авторами был разработан новый инструмент для оценки рети-нопатии недоношенных на основе количественного оценивания извилистости и расширения кровеносных сосудов. Авторы в работе [40] доказали, что извили-стость оказалась более надёжным и перспективным параметром, чем ширина сосуда для выявления РН. В работе [45] врачами исследовалась связь скорости изменения диаметра сосудов сетчатки с высоким рис-ком тяжелой ретинопатии недоношенных (РН).…”