Graefes Arch Clin Exp Ophthalmol DOI 10.1007/s00417-014-2715-2
RETINAL DISORDERS
Retinal vessel diameter measurements by spectral domain optical coherence tomography Yanling Ouyang & Qing Shao & Dirk Scharf & Antonia M. Joussen & Florian M. Heussen
Received: 8 January 2014 / Revised: 15 May 2014 / Accepted: 30 June 2014 # Springer-Verlag Berlin Heidelberg 2014
Abstract Purpose To describe a spectral domain optical coherence (OCT)-assisted method of measuring retinal vessel diameters. Methods All Patients with an OCT circle scan centered at the optic nerve head using a Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany) were retrospectively reviewed. Individual retinal vessels were identified on infrared reflectance (IR) images and given unique labels both on IR and spectral domain OCT (SD-OCT). Vessel width and vessel types obtained by IR were documented as ground truth. From OCT, measurements of each vessel, including horizontal vessel contour diameter, vertical vessel contour diameter, horizontal hyperreflective core diameter, and reflectance shadowing width, were assessed. Results A total of 220 vessels from 13 eyes of 12 patients were labeled, among which, 194 vessels (88 arteries and 65 veins confirmed from IR) larger than 40 microns were included in the study. The mean vessel width obtained from IR was 107.9±36.1 microns. A mean vertical vessel contour diameter of 119.6±29.9 microns and a mean horizontal vessel contour diameter of 124.1±31.1 microns were measured by SD-OCT. Vertical vessel contour diameter did not differ from vessel width in all subgroup analysis. Horizontal vessel contour diameter was not significantly different from vessel width for arteries and had strong or very strong correlation with vessel width for veins. Conclusion In our study, vertical vessel contour diameter measured by current commercially available SD-OCT was consistent with vessel width obtained by IR with good
Yanling Ouyang and Qing Shao contributed equally to the manuscript. Y. Ouyang : Q. Shao : D. Scharf : A. M. Joussen : F. M. Heussen (*) Department of Ophthalmology, Charité-University Medicine Berlin, Augustenburger Platz 1, 13353 Berlin, Germany e-mail: [email protected]
reproducibility. This SD-OCT based method could potentially be used as a standard measurement procedure to evaluate retinal vessel diameters and their changes in ocular and systemic disorders. Keywords Optical coherence tomography . Spectral domain . Retina vessel . Vessel diameter Abbreviations OCT SD-OCT ONH IR FA DIRC
Optical coherence tomography Spectral domain OCT Optic nerve head Infrared reflectance Fluorescein angiography Doheny Image Reading Center
Introduction Retinal blood vessels are the only visible and optically accessible small blood vessels in the human body [1]. At the same time, the vascular system is a key element in the pathophysiology of multiple systemic or localized diseases. As ophthalmologists we are provided with a unique opportunity to assess and interpret small blood vessels in live patients. We already know that this assessment offers useful parameters for the diagnosis or evaluation of ocular diseases like diabetic retinopathy and glaucoma [2]. More recently, reports have shown that quantitative assessment of retinal vessel diameters can provide independent risk information for cardiovascular or cerebral small vessel diseases, and even act as biomarkers for early detection and monitoring of Alzheimer’s disease [3–9]. Although the strength of the named correlations may be disputable, retinal vascular diameters should be assessed
Graefes Arch Clin Exp Ophthalmol
routinely and in a standardized fashion given their potential as a noninvasive cardiovascular risk prediction tool [10]. The accuracy of the measurement of retinal vessel diameters markedly influences the reliability of retinal blood flow determination. Several methods have been described for the measurement of retinal vessel diameters, most of which use retinal photography and computer-assisted image analysis to characterize and quantify subtle variations and abnormalities in the retinal vasculature [2–4, 11, 12]. In addition, fluorescein angiograms (FA) [13], indocyanine green angiography [14], retinal vessel analyzers [3, 15], and scanning laser ophthalmoscopes [14] have all been used to evaluate the diameters of retinal vessels. Optical coherence tomography (OCT) is an imaging modality capable of providing high-resolution, crosssectional images of the neurosensory retina. It has been touted to provide superior morphologic information compared with color photography and angiography [16]. Using Doppler Fourier domain OCT, in vivo human total retinal blood flow was measured [17–19]. However, for assessing retinal vessel diameter, OCT as the most often used ophthalmic imaging technique, has not been considered a useful tool. On OCT, retinal vessels appear as hyperreflective features in the inner retina [20]. The characteristics of retinal vessels have already been studied on OCT [21, 22], and automatic vessel segmentation of the optic nerve head (ONH) has been attempted in several groups with varying success [22, 23]. Given the clinical significance of measuring vessel diameters and their response to physiological stimuli, to assess vessel diameters from spectral domain, OCT (SD-OCT) is highly desirable. Our study focused on a standard SD-OCT instrument to measure retinal vessel diameters and compare these findings with the reference diameters measured from infrared images, and thus to establish a standard measurement procedure.
Methods Data collection All patients who underwent circular OCT-scans around the ONH using a Spectralis OCT+HRA (Heidelberg Engineering, Heidelberg, Germany) at our clinic between January 1, 2013 and March 1, 2013 were retrospectively collected. Eyes with normal looking retinal structure in the OCT covered area were included. Information regarding age, sex, race, history of ophthalmic diseases or surgeries, ophthalmic diagnosis, lens status, and visual acuity were also collected. Approval for data collection and analysis was obtained from the institutional review board of the Charité-University School of Medicine in Berlin. Written consent was given by the patients for their information to be stored in the hospital database and used for
research. The research adhered to the tenets set forth in the Declaration of Helsinki. The OCT scanning protocol consisted of a circular scan with 3.42–4.04 mm diameter centered on the ONH (Spectralis HRA-OCT, Heidelberg Engineering, Heidelberg, Germany). Simultaneously, near-infrared reflectance (IR) pictures were obtained. The eyes scanned with high resolution OCT mode, with a mean automatic real time of 30, and good image quality for both OCT and IR, with or without FA performed on the same day, were reviewed and analyzed using Spectralis viewing software (Heidelberg Eye Explorer, version 1.7.1.0). Grading methodology One grader (Y.O.), certified for assessing OCT and color images at the Doheny Image Reading Center (DIRC), first reviewed IR images for every eye. Using the Spectralis Software (Heidelberg Eye Explorer, version 1.7.1.0), a circle representing the OCT scan was superimposed on IR images. Vessels that overlap or share a common root with any other vessel at the intersection of the scan were excluded in the study. Any vessel that was scanned along its longitudinal axis was not included (example in Fig. 1). Each vessel included by IR was given a unique identity number (ID) for subsequent grading [24]. The corresponding vessels on OCT were labelled with the same ID. This strategy allowed each vessel from different imaging modalities (IR or OCT) to be assessed in an independent, masked fashion without knowledge of its relationship with other vessels or eye. After the initial vessel selection, two graders (Y.O., Q.S.) assessed the IR (and FA) images independently by randomly selecting the vessel ID. The classification of “Artery” or “Vein” by IR is based on the anatomical characteristics of the two vessels (e.g. shape, brightness, and central reflex) and clinical experience of the grader. For example, arteries are thinner with a wider visible central light-reflex (the lightreflex of the inner parts of the vessels) than veins. FA was only used to evaluate the vessel type when it was not obvious from IR. The identification of “Artery” by FA is based on the early dye circulation seen in the vascular loop. Vessel types determined as “Artery”, “Vein”, or “uncertain” were only marked on IR and FA images, not on OCT images. For each eye, location of the superior temporal artery, inferior temporal artery, superior temporal vein, and inferior temporal vein were also documented using the reference location shown in Fig. 2. In addition, vessel width obtained from IR (defined as minimum vessel diameter, measured vertically to the vessel axis with the calibre tool from the Spectralis software at the crossing point of vessel and OCT scanning line) was also evaluated and documented as ground truth for further analysis. The following brightness and contrast settings for IR images were used: “black on white” for color table, “medium” for sharpen, and “none” for noise reduction. Based on preliminary data in
Graefes Arch Clin Exp Ophthalmol
Fig. 1 Example for labelling vessels on infrared reflectance images and optical coherence tomography. a An optic disk fundus infrared image overlapped with one circular scan centering at the optic nerve head center. All visible retinal vessels crossing the circular scanning line were labelled with numbers starting with a random vessel. b Spectral domain optical
coherence tomography (SD-OCT) image corresponding to the circular scan visible in the infrared image. Corresponding vessels were labelled with the same number. Note: The OCT scanning line goes through the axis of Vessel Number 17 (red arrows), instead of through the vertical section of the vessel. This vessel was excluded from the analysis
our study, the color table as “black on white” offers better reproducibility for vessel width grading than “white on black” or “color” (Fig. 2). The same two graders (Y.O., Q.S.) assessed the OCT images independently by randomly selecting vessel IDs for grading. The IR (and FA) images were masked during this evaluation. From OCT, features of each vessel, including the
presence or absence of the central hyperreflective core, maximum horizontal vessel contour diameter, maximum vertical vessel contour diameter, and maximum reflectance shadowing width were assessed. These features were assessed preferably with the brightness and contrast of “black on white” for color table. If the feature of central hyperreflective core was present, its maximum horizontal diameter was also obtained,
Fig. 2 Grading examples for vessel feature measurement by optical coherence tomography (OCT). A: an artery with horizontal vessel contour diameter 124 μm (A1), vertical contour diameter 128 μm (A1) measured on the “black on white” color table; and horizontal hyperreflective core diameter 39 μm (upper in A2) and shadow width 87 μm (lower in A2) measured on “white on black” color table by
infrared images. B: a vein with horizontal vessel contour diameter 196 μm (B1), vertical contour diameter 193 μm (B1) measured on the “black on white” color table; and horizontal hyperreflective core diameter 83 μm (upper in B2) and shadow width 159 μm (lower in B2) measured on “white on black” color table by infrared images
Graefes Arch Clin Exp Ophthalmol
Fig. 3 Grading examples for vessel feature measurement by infrared reflectance (IR) image. a Grading examples for vessel locations. Superior temporal artery was labeled as 1 M (trunk), 1B (branch), or 3B (branch). Inferior temporal artery was labeled as 5 M (trunk) or 5B (branch). Superior temporal vein branches were graded as 2B and 4B. Inferior temporal vein was graded as 6 M. Additional vessels were also labeled, but were not necessarily included in the location-specific analysis. b-d
showed grading disparity of vessel width on the same vessel measured by IR choosing different color table for adjusting brightness and contrast. B: vessel width measured on “white on black” color table as 125 μm. C: vessel width measured on “black on white” color table as 83 μm. D: vessel width measured on “color” color table as 107 μm. The disparity indicated that choosing “black on white” color table offers better contrast of the retinal vessel with the background
preferably with the brightness and contrast of “white on black” for color table (examples shown in Fig. 3). In cases of disagreement, adjudication of the grading results took place to arrive at a common final answer. 16 The final agreement used as final measurement for statistical analysis. In cases where no agreement was reached, the senior grader (F.H.) made a final decision for subsequent analysis.
Results
Statistical analysis Only gradable measurements and features were used for subsequent analyses in the study. Vessel diameters less than 40 μm measured from IR were not included in the statistical analysis, because it was reported that vessels smaller than this diameter had no measurable impact on the summary values [12]. Kappa values were used to assess the inter-grader reliabilities for vessel types. The intra-reader variability for measurements of each parameter was also tested using Bland– Altman plots. Since previous publications have chosen a minimum cutting point of 85 μm for vessel trunk measurement [25], sub-analyses were also performed for vessels larger than 85 μm in diameter. Vessel width measured by IR was considered ground truth or reference measurement. The difference of each OCT parameter with the referenced vessel width was tested with a paired t-test. In addition, correlation of OCT measurements with the corresponding vessel width obtained by IR was assessed by pairwise correlation analysis (Pearson correlation). Correlation coefficients were reported with “r” and considered moderately strong for r between 0.6 and 0.8, strong for r≥0.8, and very high correlation for r≥0.9 [26, 27]. Stata (version 10.0, StataCorp LP, College Station, TX, USA) was used for the statistical analysis. A bilateral value of P
RETINAL DISORDERS
Retinal vessel diameter measurements by spectral domain optical coherence tomography Yanling Ouyang & Qing Shao & Dirk Scharf & Antonia M. Joussen & Florian M. Heussen
Received: 8 January 2014 / Revised: 15 May 2014 / Accepted: 30 June 2014 # Springer-Verlag Berlin Heidelberg 2014
Abstract Purpose To describe a spectral domain optical coherence (OCT)-assisted method of measuring retinal vessel diameters. Methods All Patients with an OCT circle scan centered at the optic nerve head using a Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany) were retrospectively reviewed. Individual retinal vessels were identified on infrared reflectance (IR) images and given unique labels both on IR and spectral domain OCT (SD-OCT). Vessel width and vessel types obtained by IR were documented as ground truth. From OCT, measurements of each vessel, including horizontal vessel contour diameter, vertical vessel contour diameter, horizontal hyperreflective core diameter, and reflectance shadowing width, were assessed. Results A total of 220 vessels from 13 eyes of 12 patients were labeled, among which, 194 vessels (88 arteries and 65 veins confirmed from IR) larger than 40 microns were included in the study. The mean vessel width obtained from IR was 107.9±36.1 microns. A mean vertical vessel contour diameter of 119.6±29.9 microns and a mean horizontal vessel contour diameter of 124.1±31.1 microns were measured by SD-OCT. Vertical vessel contour diameter did not differ from vessel width in all subgroup analysis. Horizontal vessel contour diameter was not significantly different from vessel width for arteries and had strong or very strong correlation with vessel width for veins. Conclusion In our study, vertical vessel contour diameter measured by current commercially available SD-OCT was consistent with vessel width obtained by IR with good
Yanling Ouyang and Qing Shao contributed equally to the manuscript. Y. Ouyang : Q. Shao : D. Scharf : A. M. Joussen : F. M. Heussen (*) Department of Ophthalmology, Charité-University Medicine Berlin, Augustenburger Platz 1, 13353 Berlin, Germany e-mail: [email protected]
reproducibility. This SD-OCT based method could potentially be used as a standard measurement procedure to evaluate retinal vessel diameters and their changes in ocular and systemic disorders. Keywords Optical coherence tomography . Spectral domain . Retina vessel . Vessel diameter Abbreviations OCT SD-OCT ONH IR FA DIRC
Optical coherence tomography Spectral domain OCT Optic nerve head Infrared reflectance Fluorescein angiography Doheny Image Reading Center
Introduction Retinal blood vessels are the only visible and optically accessible small blood vessels in the human body [1]. At the same time, the vascular system is a key element in the pathophysiology of multiple systemic or localized diseases. As ophthalmologists we are provided with a unique opportunity to assess and interpret small blood vessels in live patients. We already know that this assessment offers useful parameters for the diagnosis or evaluation of ocular diseases like diabetic retinopathy and glaucoma [2]. More recently, reports have shown that quantitative assessment of retinal vessel diameters can provide independent risk information for cardiovascular or cerebral small vessel diseases, and even act as biomarkers for early detection and monitoring of Alzheimer’s disease [3–9]. Although the strength of the named correlations may be disputable, retinal vascular diameters should be assessed
Graefes Arch Clin Exp Ophthalmol
routinely and in a standardized fashion given their potential as a noninvasive cardiovascular risk prediction tool [10]. The accuracy of the measurement of retinal vessel diameters markedly influences the reliability of retinal blood flow determination. Several methods have been described for the measurement of retinal vessel diameters, most of which use retinal photography and computer-assisted image analysis to characterize and quantify subtle variations and abnormalities in the retinal vasculature [2–4, 11, 12]. In addition, fluorescein angiograms (FA) [13], indocyanine green angiography [14], retinal vessel analyzers [3, 15], and scanning laser ophthalmoscopes [14] have all been used to evaluate the diameters of retinal vessels. Optical coherence tomography (OCT) is an imaging modality capable of providing high-resolution, crosssectional images of the neurosensory retina. It has been touted to provide superior morphologic information compared with color photography and angiography [16]. Using Doppler Fourier domain OCT, in vivo human total retinal blood flow was measured [17–19]. However, for assessing retinal vessel diameter, OCT as the most often used ophthalmic imaging technique, has not been considered a useful tool. On OCT, retinal vessels appear as hyperreflective features in the inner retina [20]. The characteristics of retinal vessels have already been studied on OCT [21, 22], and automatic vessel segmentation of the optic nerve head (ONH) has been attempted in several groups with varying success [22, 23]. Given the clinical significance of measuring vessel diameters and their response to physiological stimuli, to assess vessel diameters from spectral domain, OCT (SD-OCT) is highly desirable. Our study focused on a standard SD-OCT instrument to measure retinal vessel diameters and compare these findings with the reference diameters measured from infrared images, and thus to establish a standard measurement procedure.
Methods Data collection All patients who underwent circular OCT-scans around the ONH using a Spectralis OCT+HRA (Heidelberg Engineering, Heidelberg, Germany) at our clinic between January 1, 2013 and March 1, 2013 were retrospectively collected. Eyes with normal looking retinal structure in the OCT covered area were included. Information regarding age, sex, race, history of ophthalmic diseases or surgeries, ophthalmic diagnosis, lens status, and visual acuity were also collected. Approval for data collection and analysis was obtained from the institutional review board of the Charité-University School of Medicine in Berlin. Written consent was given by the patients for their information to be stored in the hospital database and used for
research. The research adhered to the tenets set forth in the Declaration of Helsinki. The OCT scanning protocol consisted of a circular scan with 3.42–4.04 mm diameter centered on the ONH (Spectralis HRA-OCT, Heidelberg Engineering, Heidelberg, Germany). Simultaneously, near-infrared reflectance (IR) pictures were obtained. The eyes scanned with high resolution OCT mode, with a mean automatic real time of 30, and good image quality for both OCT and IR, with or without FA performed on the same day, were reviewed and analyzed using Spectralis viewing software (Heidelberg Eye Explorer, version 1.7.1.0). Grading methodology One grader (Y.O.), certified for assessing OCT and color images at the Doheny Image Reading Center (DIRC), first reviewed IR images for every eye. Using the Spectralis Software (Heidelberg Eye Explorer, version 1.7.1.0), a circle representing the OCT scan was superimposed on IR images. Vessels that overlap or share a common root with any other vessel at the intersection of the scan were excluded in the study. Any vessel that was scanned along its longitudinal axis was not included (example in Fig. 1). Each vessel included by IR was given a unique identity number (ID) for subsequent grading [24]. The corresponding vessels on OCT were labelled with the same ID. This strategy allowed each vessel from different imaging modalities (IR or OCT) to be assessed in an independent, masked fashion without knowledge of its relationship with other vessels or eye. After the initial vessel selection, two graders (Y.O., Q.S.) assessed the IR (and FA) images independently by randomly selecting the vessel ID. The classification of “Artery” or “Vein” by IR is based on the anatomical characteristics of the two vessels (e.g. shape, brightness, and central reflex) and clinical experience of the grader. For example, arteries are thinner with a wider visible central light-reflex (the lightreflex of the inner parts of the vessels) than veins. FA was only used to evaluate the vessel type when it was not obvious from IR. The identification of “Artery” by FA is based on the early dye circulation seen in the vascular loop. Vessel types determined as “Artery”, “Vein”, or “uncertain” were only marked on IR and FA images, not on OCT images. For each eye, location of the superior temporal artery, inferior temporal artery, superior temporal vein, and inferior temporal vein were also documented using the reference location shown in Fig. 2. In addition, vessel width obtained from IR (defined as minimum vessel diameter, measured vertically to the vessel axis with the calibre tool from the Spectralis software at the crossing point of vessel and OCT scanning line) was also evaluated and documented as ground truth for further analysis. The following brightness and contrast settings for IR images were used: “black on white” for color table, “medium” for sharpen, and “none” for noise reduction. Based on preliminary data in
Graefes Arch Clin Exp Ophthalmol
Fig. 1 Example for labelling vessels on infrared reflectance images and optical coherence tomography. a An optic disk fundus infrared image overlapped with one circular scan centering at the optic nerve head center. All visible retinal vessels crossing the circular scanning line were labelled with numbers starting with a random vessel. b Spectral domain optical
coherence tomography (SD-OCT) image corresponding to the circular scan visible in the infrared image. Corresponding vessels were labelled with the same number. Note: The OCT scanning line goes through the axis of Vessel Number 17 (red arrows), instead of through the vertical section of the vessel. This vessel was excluded from the analysis
our study, the color table as “black on white” offers better reproducibility for vessel width grading than “white on black” or “color” (Fig. 2). The same two graders (Y.O., Q.S.) assessed the OCT images independently by randomly selecting vessel IDs for grading. The IR (and FA) images were masked during this evaluation. From OCT, features of each vessel, including the
presence or absence of the central hyperreflective core, maximum horizontal vessel contour diameter, maximum vertical vessel contour diameter, and maximum reflectance shadowing width were assessed. These features were assessed preferably with the brightness and contrast of “black on white” for color table. If the feature of central hyperreflective core was present, its maximum horizontal diameter was also obtained,
Fig. 2 Grading examples for vessel feature measurement by optical coherence tomography (OCT). A: an artery with horizontal vessel contour diameter 124 μm (A1), vertical contour diameter 128 μm (A1) measured on the “black on white” color table; and horizontal hyperreflective core diameter 39 μm (upper in A2) and shadow width 87 μm (lower in A2) measured on “white on black” color table by
infrared images. B: a vein with horizontal vessel contour diameter 196 μm (B1), vertical contour diameter 193 μm (B1) measured on the “black on white” color table; and horizontal hyperreflective core diameter 83 μm (upper in B2) and shadow width 159 μm (lower in B2) measured on “white on black” color table by infrared images
Graefes Arch Clin Exp Ophthalmol
Fig. 3 Grading examples for vessel feature measurement by infrared reflectance (IR) image. a Grading examples for vessel locations. Superior temporal artery was labeled as 1 M (trunk), 1B (branch), or 3B (branch). Inferior temporal artery was labeled as 5 M (trunk) or 5B (branch). Superior temporal vein branches were graded as 2B and 4B. Inferior temporal vein was graded as 6 M. Additional vessels were also labeled, but were not necessarily included in the location-specific analysis. b-d
showed grading disparity of vessel width on the same vessel measured by IR choosing different color table for adjusting brightness and contrast. B: vessel width measured on “white on black” color table as 125 μm. C: vessel width measured on “black on white” color table as 83 μm. D: vessel width measured on “color” color table as 107 μm. The disparity indicated that choosing “black on white” color table offers better contrast of the retinal vessel with the background
preferably with the brightness and contrast of “white on black” for color table (examples shown in Fig. 3). In cases of disagreement, adjudication of the grading results took place to arrive at a common final answer. 16 The final agreement used as final measurement for statistical analysis. In cases where no agreement was reached, the senior grader (F.H.) made a final decision for subsequent analysis.
Results
Statistical analysis Only gradable measurements and features were used for subsequent analyses in the study. Vessel diameters less than 40 μm measured from IR were not included in the statistical analysis, because it was reported that vessels smaller than this diameter had no measurable impact on the summary values [12]. Kappa values were used to assess the inter-grader reliabilities for vessel types. The intra-reader variability for measurements of each parameter was also tested using Bland– Altman plots. Since previous publications have chosen a minimum cutting point of 85 μm for vessel trunk measurement [25], sub-analyses were also performed for vessels larger than 85 μm in diameter. Vessel width measured by IR was considered ground truth or reference measurement. The difference of each OCT parameter with the referenced vessel width was tested with a paired t-test. In addition, correlation of OCT measurements with the corresponding vessel width obtained by IR was assessed by pairwise correlation analysis (Pearson correlation). Correlation coefficients were reported with “r” and considered moderately strong for r between 0.6 and 0.8, strong for r≥0.8, and very high correlation for r≥0.9 [26, 27]. Stata (version 10.0, StataCorp LP, College Station, TX, USA) was used for the statistical analysis. A bilateral value of P
