Ophthalmic Surgery, Lasers and Imaging Retina

Clinical Science 

Precise Montaging and Metric Quantification of Retinal Surface Area From Ultra-Widefield Fundus Photography and Fluorescein Angiography

Daniel E. Croft, BA; Jano van Hemert, PhD; Charles C. Wykoff, MD, PhD; David Clifton, PhD; Michael Verhoek, PhD; Alan Fleming, PhD; David M. Brown, MD

Abstract

BACKGROUND AND OBJECTIVE:

Accurate quantification of retinal surface area from ultra-widefield (UWF) images is challenging due to warping produced when the retina is projected onto a two-dimensional plane for analysis. By accounting for this, the authors sought to precisely montage and accurately quantify retinal surface area in square millimeters.

PATIENTS AND METHODS:

Montages were created using Optos 200Tx (Optos, Dunfermline, U.K.) images taken at different gaze angles. A transformation projected the images to their correct location on a three-dimensional model. Area was quantified with spherical trigonometry. Warping, precision, and accuracy were assessed.

RESULTS:

Uncorrected, posterior pixels represented up to 79% greater surface area than peripheral pixels. Assessing precision, a standard region was quantified across 10 montages of the same eye (RSD: 0.7%; mean: 408.97 mm2; range: 405.34–413.87 mm2). Assessing accuracy, 50 patients’ disc areas were quantified (mean: 2.21 mm2; SE: 0.06 mm2), and the results fell within the normative range.

CONCLUSION:

By accounting for warping inherent in UWF images, precise montaging and accurate quantification of retinal surface area in square millimeters were achieved.

[Ophthalmic Surg Lasers Imaging Retina. 2014;45:312-317.]

From the Retina Consultants of Houston (DEC, CCW, DMB), Houston, Texas; Weill Cornell Medical College, Houston Methodist Hospital (CCW, DMB), Houston, Texas; and Optos (JVH, DC, MV, AF), Dunfermline, Scotland, United Kingdom.

Drs. van Hemert, Clifton, Verhoek, and Flemming are employed by Optos. Dr. Brown is a speaker for Optos. The remaining authors have no financial or proprietary interest in the materials presented herein.

Address correspondence to Charles C. Wykoff, MD, PhD, Retina Consultants of Houston, 6560 Fannin Street, Suite 750, Houston, TX 77030; 713-524-3434; email: ccwmd@houstonretina.com.

Received: January 23, 2014
Accepted: March 25, 2014

Abstract

BACKGROUND AND OBJECTIVE:

Accurate quantification of retinal surface area from ultra-widefield (UWF) images is challenging due to warping produced when the retina is projected onto a two-dimensional plane for analysis. By accounting for this, the authors sought to precisely montage and accurately quantify retinal surface area in square millimeters.

PATIENTS AND METHODS:

Montages were created using Optos 200Tx (Optos, Dunfermline, U.K.) images taken at different gaze angles. A transformation projected the images to their correct location on a three-dimensional model. Area was quantified with spherical trigonometry. Warping, precision, and accuracy were assessed.

RESULTS:

Uncorrected, posterior pixels represented up to 79% greater surface area than peripheral pixels. Assessing precision, a standard region was quantified across 10 montages of the same eye (RSD: 0.7%; mean: 408.97 mm2; range: 405.34–413.87 mm2). Assessing accuracy, 50 patients’ disc areas were quantified (mean: 2.21 mm2; SE: 0.06 mm2), and the results fell within the normative range.

CONCLUSION:

By accounting for warping inherent in UWF images, precise montaging and accurate quantification of retinal surface area in square millimeters were achieved.

[Ophthalmic Surg Lasers Imaging Retina. 2014;45:312-317.]

From the Retina Consultants of Houston (DEC, CCW, DMB), Houston, Texas; Weill Cornell Medical College, Houston Methodist Hospital (CCW, DMB), Houston, Texas; and Optos (JVH, DC, MV, AF), Dunfermline, Scotland, United Kingdom.

Drs. van Hemert, Clifton, Verhoek, and Flemming are employed by Optos. Dr. Brown is a speaker for Optos. The remaining authors have no financial or proprietary interest in the materials presented herein.

Address correspondence to Charles C. Wykoff, MD, PhD, Retina Consultants of Houston, 6560 Fannin Street, Suite 750, Houston, TX 77030; 713-524-3434; email: ccwmd@houstonretina.com.

Received: January 23, 2014
Accepted: March 25, 2014

Introduction

Sophisticated imaging modalities are essential to managing vitreoretinal diseases. Major advances in imaging such as fluorescein angiography (FA) and optical coherence tomography (OCT) precipitated dramatic changes in the diagnosis, management, and study of retinal diseases.1,2 Recently, ultra-widefield (UWF) imaging has made it possible to resolve more than 80% of the fundus with a single capture.3 With montaging, much of the remaining retinal surface area can be resolved.4 To maximize the potential of UWF imaging, precise montaging and accurate metric quantification of retinal surface area is critical.

The Early Treatment Diabetic Retinopathy Study (ETDRS) defined a standard by which investigators could quantify retinal pathology. The posterior fundus was divided into seven standard fields, grids were overlaid on the macula to approximate area, and scales were created to grade pathology (eg, capillary loss on a scale of 0 to 4).5 These standards have been utilized in numerous clinical trials over the past 2 decades.6–8

UWF imaging captures substantially greater fundus area than isolated images of the posterior pole. Furthermore, digital imaging allows accurate and precise data analysis. These advances necessitate a new methodology for retinal quantification. It may now be possible for fundus pathology to have an objective metric: the metric system.

Currently, two platforms are capable of UWF retinal imaging. The Optos scanning laser ophthalmoscope (SLO; Optos, Dunfermline, U.K.) provides approximately 200° imaging as measured by the angle internally at the center of the eye.3 The Spectralis and HRA2 platform (Heidelberg Engineering, Heidelberg, Germany) can be used in conjunction with the Ocular Staurenghi 230 SLO contact lens (Ocular Instruments, Bellevue, WA) with an approximately 150° field of view9 or the Heidelberg noncontact UWF module with a field of view of approximately 130°.

The primary challenge to accurate quantification of retinal surface area is correcting for the warping produced when the retina, a nearly spherical surface, is projected onto a two-dimensional plane for viewing and analysis. The current study describes a standardized methodology to precisely montage UWF images and accurately quantify retinal surface area from the respective montages in square millimeters utilizing the Optos 200Tx SLO.

Methods

Montaging

To allow for precise quantification, a method was developed to standardize montages derived from five component UWF fundus images (optomaps) taken by the Optos 200Tx SLO at different directionally guided gaze angles, including on-axis, superior, inferior, nasal, and temporal images. After acquisition, each optomap image was transformed into a stereographic projection of the eye. The transformation first mapped each pixel in the optomap to a spherical 3D model eye with a diameter of 24 mm. This was performed by ray-tracing every pixel in the optomap through a combined optical model of the Optos 200Tx SLO and the Navarro UWF model eye.10 This combined optical model represented the “projection” utilized by the Optos 200Tx SLO platform to create the two-dimensional optomap from scans of a respective eye. By reversing this device-specific projection, a representation of the image data on a three-dimensional sphere was created. Thus, the result of ray tracing was a mapping table that translates pixels in the optomap to pixels on the sphere. The gaze angle of the patient, determined by the location of the fovea, was taken into account for each optomap during the ray tracing to ensure the mapping was anatomically correct. This three-dimensional model was then mapped to a two-dimensional stereographic projection by projecting all relevant pixels to a plane through the equator of the eye. The stereographic projection was conformal, hence preserving shape.

For each of the five directionally guided gaze angles, the location of the fovea was determined automatically and used to choose the mapping table for the corresponding gaze angle. Then, the four eye-steered stereographic projected images (superior, inferior, nasal, and temporal) were registered one by one to the on-axis stereographic projected image. Due to the UWF nature of the images, significant overlap occurred between quadrants. Image registration between a pair of images first extracted their vasculature and subsequently applied rotational affine translation with cross-correlation. The translation vector with the highest cross-correlation was applied to the eye-steered image in order to facilitate the overlay in the montage. Finally, segments of the original component images were joined to create a contiguous montage. A linear blend of intensity values was applied along the joins between component images.

Quantification

Following standardization of the montages for a given eye, a method to quantify retinal surface area in square millimeters was developed and automated. The most fundamental challenge in quantifying area from UWF images is accounting for the warping produced by projecting a nearly spherical surface onto a two-dimensional plane; for example, on a Mercator projection of Earth, Greenland appears to be almost the size of the continent of Africa, even though in actuality it is not. While there are a multitude of different mathematical projections available to preserve various attributes of a three-dimensional object, the azimuthal projection is the prototypical choice used to digitally project the fundus.11 In azimuthal projections, image distances from a central point are calculated by a function of true distance independent of angle, preserving directionality from a central point, the posterior pole. By preserving directionality, vascular landmarks extending radially into the periphery are portrayed most accurately. However, this is at the cost of distorting size and shape proportionally to increasing radial distance. Thus, warping becomes substantially more pronounced further into the periphery.

To correct for warping in area measurements, the pixels intended to be quantified were first mapped from the montaged two-dimensional stereographic projection to the three-dimensional spherical representation referenced previously. Any pair of points on a sphere defines a great circle on that sphere, and a set of at least three great circles defines an area (Figure 1, page 313). To quantify the area denoted by a polygon (ie, any area selected in pixels), the surface area (S) was calculated by taking the sum of the radian angles (θ) of the intersecting great circles on a sphere with a radius (R = 12 mm) through the vertices that define the polygon (of n sides) and applying the following well-established equation:12

S=[θ−(n−2)π] R2

(A) Stereographic representation of a single optomap fluorescein angiogram (not a montage) with annotations. Three intersecting red lines define the area to be quantified (pink triangle). (B) Three-dimensional spherical representation of the same image and annotations. The polygon depicted (pink triangle) is bounded by three great circles (n=3).

Figure 1.

(A) Stereographic representation of a single optomap fluorescein angiogram (not a montage) with annotations. Three intersecting red lines define the area to be quantified (pink triangle). (B) Three-dimensional spherical representation of the same image and annotations. The polygon depicted (pink triangle) is bounded by three great circles (n=3).

To assess the precision of quantification, one of the researchers (DEC) performed several FAs over a 2-month period of an anatomically stable, healthy eye. A montage was produced from each of these FA sessions, and four vascular landmarks were selected, one in each quadrant of the completed montage. A line was drawn between each of these four landmarks to create a standardized quadrilateral. The area inside this quadrilateral was subsequently quantified in each montage. This method was chosen so that areas from each component image and their blended regions would be included in the quantification calculations, allowing for a more holistic measurement of precision. The relative standard deviation of the areas was then calculated, and an analysis of variance was performed.

To assess the accuracy of quantification, the optic nerve head disc area from 50 patients’ montages was measured ( ClinicalTrials.gov identifiers: NCT01552408 and NCT01710839). The mean and range of these disc area calculations were compared to published normative disc area measurements by Samarawickrama et al.13

To assess the warping inherent in montaged UWF images, one disc area was defined by the circle of pixels constituting the optic disc in a representative montage. This circle of pixels was then translated radially outward from its original location at the optic disc, toward the periphery. As it was translated, the circle was periodically (every 50 pixels) quantified in square millimeters until reaching the ora serrata.

Clinical Applications

To quantify nonperfusion in a patient with diabetic retinopathy, a UWF FA montage was cropped to create a standard field, which was centered on the posterior pole. This field was maximized to include as much gradable fundus as possible. Two researchers (DEC, CCW) used the “magic wand” tool in Photoshop CS6 to select regions of nonperfusion; an edge-detection algorithm, the tool allows manual definition of a tolerance range of pixel intensities and selection of a pixel in a region of nonperfusion. This selected pixel serves as the baseline pixel intensity for the selection tool. The algorithm then identifies surrounding pixels in the range ± the defined tolerance. The selection propagates outward contiguously, with a sensitivity defined by the tolerance, until the region of nonperfusion is selected. The larger the tolerance, the further the selection will propagate with each selection. Vasculature and regions of retinal perfusion are brighter (greater pixel intensity), thus serving as barriers to the selection. Another customizable option for the magic wand tool that was utilized was 3 × 3 averaging: upon clicking in a region of nonperfusion, the surrounding pixel’s intensities are averaged to create the baseline intensity employed by the selection algorithm. Once all areas of nonperfusion were selected, they were quantified as above.

Results

To assess precision, a standard region demarcated by four vascular landmarks was quantified across 10 unique montages, from 10 unique fluorescein angiograms of the same eye (see Methods section). The mean number of pixels selected within the quadrilateral was 1,698,425 (range: 1,727,700–1,680,920 pixels; SD: 14,032 pixels). Thus, the standard deviation was 0.8% of the mean, reflecting a high degree of precision (Figure 2). The mean area of the quadrilateral quantified across the 10 unique FA montages was 408.97 mm2 (range: 405.34–413.87 mm2; SD: 2.91 mm2; variance: 8.49 mm2). Thus, the standard deviation was 0.7% of the mean, again reflecting a high degree of precision.

Method utilized to assess the precision of montaging and quantification. Four vascular landmarks define the vertices of the red quadrilateral, which was quantified across 10 unique fluorescein angiography montages from the same anatomically stable, healthy eye.

Figure 2.

Method utilized to assess the precision of montaging and quantification. Four vascular landmarks define the vertices of the red quadrilateral, which was quantified across 10 unique fluorescein angiography montages from the same anatomically stable, healthy eye.

To assess accuracy, the mean disc area of 50 patients was quantified and compared to a published standard of normative disc area measurements with a range of 1.60 to 2.63 mm2 mean disc areas.13 The mean disc area quantified from 50 patients in the current work fell within the published accepted range at 2.21 mm2 (SE: 0.06 mm2; range: 1.25–3.16 mm2).

After confirming the precision and accuracy of the methodology described above, the warping inherent in UWF imaging was assessed. The optic disc in a selected montage was determined to have a radius of 49 pixels and include a total of 7,667 pixels, defining one disc area. In its native location, these pixels represented a disc area of 2.35 mm2. This circle of pixels was translated radially toward the periphery and quantified every 50 pixels (Figures 3A–B). At the anatomic posterior pole (fovea), this circle of pixels represented a disc area of 2.48 mm2. After crossing the posterior pole, each circle of pixels represented less surface area the further into the periphery it was located. Specifically, this circle of pixels represented 1.39 mm2 and 0.50 mm2 after being translated radially by 1,000 pixels and 1,700 pixels, respectively, a difference of 41% and 79% from the actual disc area (Figure 3B). The ora serrata was estimated to begin 1,700 pixels radially from the temporal edge of the optic nerve head.

To demonstrate warping inherent in ultra-widefield (UWF) imaging, one disc area was defined by the number of pixels constituting the optic disc. (A) This circle of pixels was translated radially from its original location into the temporal periphery and quantified every 50 pixels as illustrated representatively by red circles. (B) A graph illustrating the warping measured across the UWF image. After crossing the posterior pole (yellow dot), each pixel represents less surface area the further into the periphery it is located. The ora serrata was estimated to begin 1,700 pixels radially from the temporal edge of the optic nerve head.

Figure 3.

To demonstrate warping inherent in ultra-widefield (UWF) imaging, one disc area was defined by the number of pixels constituting the optic disc. (A) This circle of pixels was translated radially from its original location into the temporal periphery and quantified every 50 pixels as illustrated representatively by red circles. (B) A graph illustrating the warping measured across the UWF image. After crossing the posterior pole (yellow dot), each pixel represents less surface area the further into the periphery it is located. The ora serrata was estimated to begin 1,700 pixels radially from the temporal edge of the optic nerve head.

To quantify nonperfusion in a patient with diabetic retinopathy, a standard field was defined as an ellipse covering an area of 730 mm2 (282 disc areas; 3,954,040 pixels) centered on the optic disc. The patient had 195 mm2 (1,263,090 pixels) of nonperfusion or an ischemic index of 26.7% (nonperfused / total retinal surface area: 195/730 mm2). Left uncorrected for retinal concavity, the ischemic index calculated by pixel ratio would be 32.0%; this numerical difference of 5.3% represents a 19.9% error (Figures 4A–B, page 316). This error is magnified the larger and more peripheral the retinal area under consideration (Figure 4C); for example, when the far periphery is isolated and quantified (1,625,900 pixels; area: 218 mm2), it accounts for 29.9% (218/730 mm2) of the standard field. Calculated by pixel ratio, it accounts for 41.2% of the standard field, a numerical difference of 11.3%, which represents a 38.0% error.

Clinical example quantifying retinal nonperfusion in a patient with diabetic retinopathy. (A) Standard field defined to quantify nonperfusion. (B) Selection of area of retinal nonperfusion (red). (C) Selection of peripheral retina (red).

Figure 4.

Clinical example quantifying retinal nonperfusion in a patient with diabetic retinopathy. (A) Standard field defined to quantify nonperfusion. (B) Selection of area of retinal nonperfusion (red). (C) Selection of peripheral retina (red).

Discussion

ETDRS introduced the first standard for quantifying retinal pathology, which has endured for over 2 decades. Its endurance is a testament to its effective design. With ETDRS grids and scales, reading centers can grade images relatively quickly with a high degree of reproducibility and intergrader agreement. Area is calculated in relation to disc area, whereas an average optic disc diameter was originally defined as 1,500 μm. The grid is 4 disc diameters wide and covers a total of 16 disc areas.5 This was an acceptable method of quantification for the 30° photos that ETDRS utilized of the posterior pole; with a narrow field of view, it was not necessary to account for the concavity of the retinal surface (Figure 5).

Ultra-widefield montage field of view compared to the standard ETDRS grid used for retinal quantification.

Figure 5.

Ultra-widefield montage field of view compared to the standard ETDRS grid used for retinal quantification.

With the introduction of UWF imaging, new light has been shed on common retinal pathologies. For example, peripheral nonperfusion is now readily apparent in many eyes with retinal vascular disease. Several investigators have suggested that an ischemic index (ISI) can be calculated in eyes with retinal vein occlusion by quantifying retinal nonperfusion in order to provide a metric for disease severity, potentially serving as a prognostic indicator and predictor of complications.4,14,15 Initial attempts to calculate an ISI with both Staurenghi and Optos C200 SLO UWF images had limited success. Nonperfused regions were selected to determine the number of “ischemic pixels” in a defined field. This value was then divided by the total number of pixels in the field, effectively calculating the percentage of nonperfused pixels.14,15 However, these approaches did not address the warping inherent in UWF images. Furthermore, these approaches did not quantify retinal surface area in an absolute metric, square millimeters.

Addressing the issue of warping due to retinal concavity, RF Spaide elegantly pioneered a method to montage and quantify UWF FA images from the Optos P200 SLO4. Briefly, Spaide used an on-axis UWF image of the posterior pole as a foundation to subsequently transform peripherally guided images around with elastic deformation. He manually defined control points and utilized thin-plate splines to align each peripheral image with the on-axis image. The montage was then warped to fit a schematic anatomical model retina and quantified. However, defining manual control points introduces a source of operator variance. The transformations and montaging operations of the current methodology were designed to be fully automated so as to produce the same result given the same input. Furthermore, by not applying a transformation to the on-axis image, the projection used by Spaide was not defined, making conversions to square millimeters less accurate. Taking a different approach, the current montaging and quantification methods were based on optical modeling of the imaging platform and eye.

The current report encountered limitations that warrant further investigation and development. Utilizing a model eye with a diameter of 24 mm for optical modeling introduces error due to the natural variance of individual ocular dimensions. Furthermore, our assessments of precision and accuracy required the manual definition of regions to be quantified, introducing a source of operator variance. Finally, assessing the accuracy of peripheral retinal measurements is challenging because there are no accepted standards for the dimensions of peripheral retinal landmarks. True accuracy is problematic to assess without in vivo or postmortem measurements to correlate with fundus images.

With an effective method to accurately and precisely quantify retinal surface area from UWF images, many clinical and research applications may be realized. The current report demonstrates the ability to quantify the area of retinal nonperfusion in square millimeters. Other applications may include measuring the area of retinal detachment, hemorrhage, and retinal or choroidal lesions, as well as tracking laser treatments and progression of geographic atrophy. However, many challenges remain before the ETDRS standard can be substituted. With a substantially larger area to analyze, an ergonomic method to efficiently and accurately select regions for quantification is needed. Furthermore, well-defined procedures are required to maximize reproducibility and intergrader agreement. The challenges associated with intergrader and even intragrader agreement are substantial when quantifying retinal nonperfusion; selecting individual pixels that represent scattered and diffuse regions of retinal nonperfusion from a gray-scale, high-resolution montage of component images captured at slightly different angiographic phases creates the possibility for large variance between readers and even over time for the same reader.

Modern UWF imaging modalities produce a nearly complete retinal portrait and allow for sophisticated digital analysis. However, a UWF fundus view also introduces significant warping when it is projected onto a two-dimensional surface for viewing. Precise montaging and accurate quantification of retinal surface area that addresses this warping is essential to realizing the full potential of UWF imaging.

In the current report, a relative standard deviation of 0.7% when assessing the area of a defined quadrilateral across 10 unique montages from the same eye reflects a high degree of precision. The quantification methodology was accurate because the mean disc area quantified across 50 patients fell within the range of established measurements. Lastly, the dramatic peripheral warping inherent in UWF images underscores the necessity of correcting for retinal concavity. When quantifying peripheral retinal pathology without accounting for retinal concavity, the error can be substantial.

References

  1. Novotny HR, Alvis DL. A method of photographing fluorescence in circulating blood in the human retina. Circulation. 1961;24:82–86. doi:10.1161/01.CIR.24.1.82 [CrossRef]
  2. Huang D, Swanson EA, Lin CP, et al. Optical coherence tomography. Science. 1991;254(5035):1178–1181. doi:10.1126/science.1957169 [CrossRef]
  3. Manivannan A, Plskova J, Farrow A, Mckay S, Sharp PF, Forrester JV. Ultra-wide-field fluorescein angiography of the ocular fundus. Am J Ophthalmol. 2005;140(3):525–527. doi:10.1016/j.ajo.2005.02.055 [CrossRef]
  4. Spaide RF. Peripheral areas of nonperfusion in treated central retinal vein occlusion as imaged by wide-field fluorescein angiography. Retina. 2001;31(5):829–837. doi:10.1097/IAE.0b013e31820c841e [CrossRef]
  5. Early Treatment Diabetic Retinopathy Study Research Group. Classification of diabetic retinopathy from fluorescein angiograms. Ophthalmology. 1991;98(5 Suppl):807–822. doi:10.1016/S0161-6420(13)38013-0 [CrossRef]
  6. Blodi BA, Domalpally A, Scott IU, et al. Standard care vs corticoste-roid for retinal vein occlusion (SCORE) Study system for evaluation of stereoscopic color fundus photographs and fluorescein angiograms: SCORE study report 9. Arch Ophthalmol. 2010;128(9):1140–1145. doi:10.1001/archophthalmol.2010.193 [CrossRef]
  7. Brown DM, Kaiser PK, Michels M, et al. Ranibizumab versus verte-porfin for neovascular age-related macular degeneration (ANCHOR). N Engl J Med. 2006;355(14):1432–1444. doi:10.1056/NEJMoa062655 [CrossRef]
  8. Rosenfeld PJ, Brown DM, Heier JS, et al. Ranibizumab for neovas-cular age-related macular degeneration (MARINA). N Engl J Med. 2006;355:1419–1431. doi:10.1056/NEJMoa054481 [CrossRef]
  9. Staurenghi G, Viola F, Mainster MA, et al. Scanning laser ophthalmoscopy and angiography with a wide-field contact lens system. Arch Ophthalmol. 2005;123(2):244–252. doi:10.1001/archopht.123.2.244 [CrossRef]
  10. Escudero-Sanz I, Navarro R. Off-axis aberrations of a wide-angle schematic eye model. J Opt Soc Am A Opt Image Sci Vis. 1999;16(8):1881–1891. doi:10.1364/JOSAA.16.001881 [CrossRef]
  11. Bugayevskiy LM, Snyder JP. Map Projections: A Reference Manual. London, UK: CRC Press; 1995.
  12. Donnay JDH. Spherical Trigonometry. New York, NY: Interscience Publishers; 2007.
  13. Samarawickrama C, Hong T, Jonas JB, et al. Measurement of normal optic nerve head parameters. Surv Ophthalmol. 2012;57(4):317–336. doi:10.1016/j.survophthal.2011.12.001 [CrossRef]
  14. Patel RD, Messner LV, Teitelbaum B, et al. Characterization of ischemic index using ultra-widefield fluorescein angiography in patients with focal and diffuse recalcitrant diabetic macular edema. Am J Ophthalmol. 2013;155(6):1038–1044. doi:10.1016/j.ajo.2013.01.007 [CrossRef]
  15. Tsui I, Kaines A, Havunjian MA, et al. Ischemic index and neovascularization in central retinal vein occlusion. Retina. 2011;31:105–110. doi:10.1097/IAE.0b013e3181e36c6d [CrossRef]

10.3928/23258160-20140709-07

Sign up to receive

Journal E-contents