Three-Dimensional Optical Coherence Tomography (3D-OCT …

نوشته شده در موضوع عکسهای سه بعدی خیابانی در ۱۸ آذر ۱۳۹۳

  1. Hiroshi Ishikawa
    ۱
    ,
    ۲
    ,
  2. Jongsick Kim
    ۱
    ,
    ۲
    ,
  3. Thomas R. Friberg
    ۱
    ,
    ۲
    ,
  4. Gadi Wollstein
    ۱
    ,
  5. Larry Kagemann
    ۱
    ,
    ۲
    ,
  6. Michelle L. Gabriele
    ۱
    ,
    ۲
    ,
  7. Kelly A. Townsend
    ۱
    ,
  8. Kyung R. Sung
    ۱
    ,
  9. Jay S. Duker
    ۳
    ,
  10. James G. Fujimoto
    ۴
    and
  11. Joel S. Schuman
    ۱
    ,
    ۲

  1. ۱From a UPMC Eye Center, Eye and Ear Institute, Ophthalmology and Visual Science Research Center, Department of Ophthalmology,
    University of Pittsburgh School of Medicine, Pittsburgh, Pennsylvania; the

  2. ۲Department of Bioengineering, Swanson School of Engineering, University of Pittsburgh, Pittsburgh, Pennsylvania;

  3. ۳New England Eye Center, Tufts Medical Center, Tufts University School of Medicine, Boston, Massachusetts; and the

  4. ۴Department of Electrical Engineering and Computer Science and Research Laboratory of Electronics, Massachusetts Institute
    of Technology, Cambridge, Massachusetts.

Abstract

purpose. To rise a semiautomated routine to daydream structures of seductiveness (SoIs) along their contour within three-dimensional,
bright domain visual conformity tomography (3D SD-OCT) data, though a need for segmentation.

methods. With a use of dual SD-OCT devices, a authors achieved 3D SD-OCT information within 6 × ۶ × ۱٫۴-mm and 6 × ۶ × ۲-mm volumes, respectively,
centered on a fovea in healthy eyes and in eyes with retinal pathology. C-mode images were generated by sampling a variable
density craft semiautomatically modeled to fit a contour of a SoI. Unlike published and commercialized methods, this
routine did not need retinal covering segmentation, that is famous to destroy frequently in a participation of retinal pathology.
Four SoIs were visualized for healthy eyes: warp of retinal haughtiness fiber (RNF), retinal capillary network (RCN), choroidal
capillary network (CCN), and vital choroidal vasculature (CV). Various SoIs were visualized for eyes with retinal pathology.

results. Seven healthy eyes and 7 eyes with retinal pathology (cystoid macular edema, executive serous retinopathy, vitreoretinal
traction, and age-related macular degeneration) were imaged. CCN and CV were successfully visualized in all eyes, whereas
RNF and RCN were visualized in all healthy eyes and in 42.8% of eyes with pathologies. Various SoIs were successfully visualized
in all eyes with retinal pathology.

conclusions. The due C-mode contour displaying might yield clinically useful images of SoIs even in eyes with serious pathologic changes
in that segmentation algorithms fail.

Medical imaging has prolonged played a vicious purpose in diagnosing and assessing several pathologic conditions. In a past decade
alone, along with a quick expansion of mechanism technologies, several new imaging modalities have been introduced. These
yield three-dimensional (3D) picture datasets and are proof themselves to be clinically useful. One of a categorical advantages
afforded with a 3D picture dataset is stretchable picture cognisance capability. One can cut a aim hankie in any arbitrary
plane, permitting clinicians to investigate abnormalities from several vantage points (e.g., sagittal, coronal, and craft sections).

Optical conformity tomography (OCT) was grown in 1991 and was commercially introduced to ophthalmology in 1996.12 First-generation OCT instruments supposing a 2D cross-sectional viewpoint of a tellurian retina during a little fortitude (10
μm) in vivo. Because of a noncontact and noninvasive imaging capabilities, OCT quick became an indispensable clinical
apparatus for assessing retinal abnormalities, including a heading causes of blindness: age-related macular degeneration, diabetic
retinopathy, and glaucoma.2345 New techniques formed on spectral/Fourier domain display have achieved approximately 50× increases in imaging speeds.678 With a quick axial imaging rate, SD-OCT has enabled us to obtain comparatively high-density 3D brick information of a tellurian retina
within a reasonable prove time (e.g., 27 kHz sampling or 200 × ۲۰۰ × ۱۰۲۴ samples in approximately 1.48 seconds; Fig. 1a ). Although compulsory C-mode images along a planes perpendicular to a scanning pivot yield an engaging perspective,
interpretation of such C-mode images is not discerning since mixed opposite retinal layers are sliced in a same view
given a healthy round span of a eye (Fig. 1b) .910 In serve to this healthy hankie curvature, an eye moves along a z-axis on a little scale even within a brief scanning time, heading to a twisted 3D SOCT picture dataset (Fig. 1c) .

Figure 1.

(a) SD-OCT representation images display (left) craft cut and (right) round slice. (b) Conventional C-mode viewpoint of tellurian retina achieved by antecedent SD-OCT territory (501 × ۱۸۰ × ۱۰۲۴ samplings in a 6 × ۶ × ۱٫۴-mm
region). C-mode picture (left) shows a concentric round hardness and a twisted viewpoint of a aim covering structure that are caused by rupturing multiple
opposite covering structures in a retina with a true line (light blue band, right). (c) A 3D SD-OCT information set that might seem to have minimal eye fit on observation of a craft territory (left) shows poignant axial eye fit along a straight territory (right).

A blurb SD-OCT (Cirrus HD-OCT; Carl Zeiss Meditec, Dublin, CA; module chronicle 3.0) was used to solve this problem because
it provides a capability of sampling a OCT information set along a segmented covering (internal tying surface [ILM], retinal
colouring epithelium [RPE], or propitious bend to a RPE) with capricious thickness. This segmented C-mode routine contains only
structures in a segmented layer(s) and provides an softened viewpoint compared with compulsory C-mode, that intersects multiple
hankie planes (Fig. 2) . Unfortunately, a peculiarity of a segmented C-mode picture depends on a peculiarity of a programmed segmentation algorithm
performance. The decrease of segmentation algorithm opening in a participation of retinal pathology is good described
(Fig. 3) .1112

Figure 2.

OCT segmented C-mode image. C-mode picture (left) shows a projection from a ILM (right, within light blue lines) with a density of 23 μm.

Figure 3.

Segmentation blunder sample. The segmented ILM limit (top, light blue line) is not following a ILM scrupulously during a cystic lesion (arrows) since of surprising retinal structure and bad vigilance quality. The smoothed propitious line (magenta) of a RPE also shows a hankie limit clarification that has no association with a rescued RPE limit (blue line).

The purpose of this investigate was to rise a routine to simply besiege structures of seductiveness (SoIs) while generating C-mode
images in a semiautomated conform and to report a clinical applications of such a technique.

Methods

Seven healthy eyes and 7 eyes with retinal pathology (cystoid macular edema, executive serous retinopathy, vitreoretinal
traction, age-related macular degeneration) were enrolled by retrospectively reviewing a scanning record during a University
of Pittsburgh Medical Center Eye Center. Healthy eyes had no story or justification of intraocular surgery, retinal pathology
or glaucoma, or refractive errors of reduction than 8 D and normal-appearing ONHs. All healthy eyes had normal formula of comprehensive
visible hearing and programmed perimetry commentary within normal limits.

The macular segment was scanned on all eyes regulating possibly of dual SD-OCT devices, an ultrahigh-resolution investigate prototype
of a possess settlement and Cirrus OCT (Carl Zeiss Meditec). The ultrahigh-resolution investigate antecedent had 3.0- to 3.5-μm axial
picture fortitude and an imaging speed of 25,000 axial scans per second, since a Cirrus OCT (Carl Zeiss Meditec) had 5-μm
fortitude and a speed of 27,000 axial scans per second. All imaging was achieved by undilated pupils. Institutional
examination house and ethics cabinet capitulation were achieved for a study, and sensitive agree was achieved from all subjects.
This investigate followed a beliefs of a Declaration of Helsinki and was conducted in correspondence with a Health Insurance Portability
and Accountability Act.

Image Acquisition

Prototype SD-OCT.

Three-dimensional brick OCT information were achieved with a raster-scanning settlement centered during a fovea and consisting of 180 frames
of craft linear B-scan with 501 A-scan lines (501 × ۱۸۰ × ۱۰۲۴ [width × tallness × depth] voxels within 6 × ۶ × ۱٫۴ mm,
prove time 3.84 seconds). Images with bad vigilance turn (subjectively assessed) were discarded. In addition, images with detectable
eye fit via a scan—larger than one vessel hole or a vital exaggeration of a foveal segment in OCT fundus images—were
discarded. Raw bright domain vigilance information were exported and processed into concept time-domain power information format with
a module module of a possess design.

Cirrus OCT.

Three-dimensional brick OCT information were achieved regulating a macular brick 200 × ۲۰۰ prove pattern, that consists of 200 frames
of a craft linear B-scan with 200 A-scan lines (200 × ۲۰۰ × ۱۰۲۴ voxels within 6 × ۶ × ۲ mm; prove time, 1.48 seconds).
Images with vigilance strength (SS) reduction than 8 were deliberate of bad peculiarity and were discarded. In addition, images with
detectable eye fit were rejected formed on a same OCT fundus picture criteria described. OCT prove tender information (processed by
Cirrus complement software, chronicle 3.0; Carl Zeiss Meditec) were exported and processed into concept time-domain intensity
information format with a same module to safeguard that 3D brick OCT information achieved with possibly of a inclination were serve processed
in a same manner.

C-Mode Image Generation.

C-mode images were generated by a contour displaying method, sampling a non-static density craft semiautomatically modeled to
fit a contour of a SoI. This interactive, semiautomated routine was achieved with a module module of a possess design,
as described below:

  1. User places mixed anchor points, that are connected with a spline-interpolated line, on one of a craft sections
    (Fig. 4) . Each user submit (moving/deleting an existent anchor prove or adding a new anchor point) triggers a C-mode picture update,
    providing real-time communication to a user for easy manipulation.

  2. User repeats step 1 on a straight territory (Fig. 5) .

  3. User repeats stairs 1 and 2 to get a acceptable outcome contour-modeled C-mode image. Modeled contours in craft and
    straight sections are interlinked so that changes can be done in any territory though are automatically and instantly applied
    to a whole image. More anchor points are typically indispensable on a straight territory since of eye fit along a z-axis
    during imaging (Fig. 6) . At a end, a formidable 3D contour indication of SoI is generated internally.

Figure 4.

Cystic changes in a macular region. Multiple layers are represented in a initial C-mode picture (top left), creation interpretation difficult. After adjusting a locations of 3 anchor points on a singular B-mode tomogram (horizontal section, bottom right; red line on left defines a retinal plcae of a singular B-mode tomogram on a right), a C-mode picture (bottom left) starts to uncover a some-more accurate image. Note a variability of power along several horizontal frames since of z-axis equivalent irregularities in a vertical section.

Figure 5.

Switching to a vertical territory (along a red line, left), a C-mode picture (top) stays a same as a C-mode picture from Figure 4 (bottom) since no contour displaying on a straight territory has been performed. Note a angled contour on a straight section
caused by eye fit along z-axis during imaging. After adjusting 3 anchor locations in a singular B-mode tomogram (bottom right), many of a craft artifact on a C-mode picture is resolved (bottom left).

Figure 6.

Further enlightening a straight contour displaying regulating mixed anchor points suggested a full border of a cystic changes.

Image Evaluation,

Four SoIs were targeted for visualization: warp of retinal haughtiness fiber (RNF), retinal capillary network (RCN), choroidal
capillary network (CCN), and vital choroidal vasculature (CV). C-mode images for any of these 4 SoIs were generated with
compulsory (flat plane), segmented (conventional adaptive threshold segmentation of ILM, RNFL, and RPE),11 and contour-modeling methods. Incomplete sketchy cognisance of an SoI was deliberate a failure. In addition, several SoIs
(e.g., cystoid lesion, drusen) were visualized for eyes with retinal pathologies.

Results

CCN and CV were successfully visualized in all eyes; RNF and RCN were visualized in 100% of healthy eyes and 42.8% of eyes
with abnormalities (Figs. 7 8 9 10 , generated from a same 3D brick information acquired regulating a antecedent SD-OCT). Primary reasons for catastrophic visualization
of RNF and RCN in eyes with abnormalities were unsuitable vigilance turn and difficult-to-model contour of a infirm SoIs.
Conventional C-mode images showed mixed layers in a same slice, creation it harder to sense a aim structure than
segmented and contour-modeling images. Although there was no biased disproportion in picture peculiarity between segmented and
contour-modeling images for CCN and CV (Figs. 9 10) , a contour-modeling images visualized SoIs improved did than a segmented images for RCN since a contour line was different
from that of ILM and RPE (Fig. 8) . The contour-modeling routine also showed improved RNF images than did a segmented routine (Fig. 7) .

Figure 7.

RNF striation. Conventional C-mode (top) is injured with a reduction of mixed structures (dark area represents a translucent cavity). Segmented C-mode (middle) shows a warp nearby a fovea though is not as minute as a contour displaying C-mode (bottom). Note that this center picture was achieved with a antecedent SD-OCT territory and that a picture for a segmented C-mode was
initial flattened to a segmented ILM, ensuing in a twisted cross-sectional image.

Figure 8.

RCN. Conventional C-mode (top) is injured by a reduction of mixed structures. Segmented C-mode (middle) shows a capillary network nearby a fovea, though other layers are visualized in a marginal area. Contour-modeling C-mode
(bottom) shows a many finish viewpoint of a RCN. This picture was achieved with a antecedent SD-OCT territory and was initial flattened
to a segmented RPE for both segmented and contour-modeling C-mode.

Figure 9.

CCN. Conventional C-mode (top) is injured by a reduction of mixed structures. Segmented and contour-modeling C-mode images (bottom) uncover a capillary network clearly. This picture was achieved with a antecedent SD-OCT territory and was initial flattened to the
segmented RPE for segmented and contour-modeling C-mode.

Figure 10.

CV. Conventional C-mode (top) shows CV in a executive area though not in a marginal area. Segmented and contour-modeling C-mode images (bottom) uncover CV clearly. This picture was achieved with a antecedent SD-OCT territory and was initial flattened to a segmented RPE for
segmented and contour-modeling C-mode.

Various SoIs were successfully visualized in all eyes with retinal abnormalities that could not be entirely comprehended by examining
particular cross-sectional images (Figs. 11 12 13 14) . A theme with cystoid macular edema had a cystoid lesion with an engaging vascular network (Fig. 11) . Another with executive serous retinopathy had serous unconcern in a foveal segment with a rarely contemplative structure,
expected analogous to a segment of steam causing a unconcern (Fig. 12) . A theme with soppy ARMD had transparent division lines of mixed drusen with noted betterment of a fovea (Fig. 13) . An eye with estimable vitreoretinal traction had retinal folds during a turn of a ILM centered during a traction point
(Fig. 14) .

Figure 11.

Cystoid macular edema. Contour-modeling C-mode reveals a inner structure of a cystoid lesion compared with possible
neovascular arrangement during a temporal side.

Figure 12.

CSR. Contour-modeling C-mode shows a serous unconcern in a foveal segment and a rarely contemplative structure inferior-nasal
to a fovea, expected analogous to a segment of steam as demonstrated on shimmer angiography (FAG image, inset).

Figure 13.

ARMD. Contour-modeling C-mode shows clearly demarcated mixed drusen. Centrally, a drusenoid colouring epithelium detachment
is found.

Figure 14.

Vitreoretinal traction delegate to an intraocular infection months earlier. Contour-modeling C-mode shows retinal folds at
a ILM turn centered during a prove of biggest traction. Note that a cross-sectional OCT picture (right) is sampled along a straight line (left, red line) to daydream a retinal folds during a ILM turn clearly.

Discussion

The ideal routine for displaying C-mode images is to uncover usually a aim covering or structure. Using a segmentation information
should, therefore, be a healthy trail heading to ideal C-mode images. However, as described, automatically segmenting pathologic
structures is still a formidable task. This generates a need for an halt resolution to simply daydream specific SoIs to
take full advantage of a information embedded in 3D OCT information sets.

The benefaction formula uncover a transparent advantage of contour-modeling C-mode over compulsory C-mode. All nonspecific SoIs, except
RCN, were visualized equally good with contour-modeling and segmented C-mode since ILM and RPE segmentation worked well
on healthy samples. RCN has a contour line opposite adequate from that of a ILM and a RPE that contour-modeling C-mode
outperformed a segmented C-mode even though segmentation failure.

In subjects with abnormalities, ILM and RPE segmentation unsuccessful frequently, that done segmented C-mode images formidable to
interpret. On a other hand, a contour-modeling routine supposing consistently good-quality C-mode images, even with complicated
pathologic structures.

A reduction of contour displaying is a requirement for user interaction, though this proceed is a viable resolution given the
stream standing of programmed segmentation. This routine typically requires usually 3 or 4 anchor points in a horizontal
territory since craft raster prove sections are acquired some-more quick and uncover reduction eye motion-related exaggeration than
straight sections. Ten to 30 anchor points are typically compulsory in straight sections, depending on a magnitude of z-direction eye fit during imaging. Given that a contour line between a anchors is automatically smoothed by spline
interpolation and a C-mode picture is updated in genuine time (fully interactive), a altogether time compulsory for a lerned operator
to routine one 3D OCT dataset is 1 to 2 minutes.

When segmentation is successful, it is easier to start from a segmented craft than a prosaic craft and to serve cgange the
contour indication to fit a clinician’s needs. Having a hybrid choice might promote faster and easier processing. This feature
has nonetheless to be implemented and evaluated.

In conclusion, a due contour-modeling C-mode routine might yield clinically useful images of intraretinal structures
even in eyes with serious pathologic changes in that segmentation algorithms fail.

Footnotes

  • Presented in partial during a annual assembly of a Association for Research in Vision and Ophthalmology, Fort Lauderdale, Florida,
    Apr 2008.

  • Supported in partial by National Institutes of Health contracts R01-EY13178–۰۸, R01-EY11289–۲۳, and P30-EY08098–۲۲, The Eye and
    Ear Foundation (Pittsburgh, PA), and unlimited grants from Research to Prevent Blindness, Inc. (New York, NY).

  • Submitted for announcement Aug 11, 2008; revised Oct 3, 2008; supposed Jan 15, 2009.

  • Disclosure: H. Ishikawa, None; J. Kim, None; T.R. Friberg, None; G. Wollstein, Carl Zeiss Meditec (F), Optovue (F); L. Kagemann, None; M.L. Gabriele, None; K.A. Townsend, None; K.R. Sung, None; J.S. Duker, None; J.G. Fujimoto, Optovue (I, C), Carl Zeiss Meditec (P); J.S. Schuman, Carl Zeiss Meditec (P)

  • The announcement costs of this essay were defrayed in partial by page assign payment. This essay contingency therefore be marked
    advertisement” in suitability with 18 U.S.C. §۱۷۳۴ only to prove this fact.

  • Corresponding author: Gadi Wollstein, UPMC Eye Center, Eye and Ear Institute, Ophthalmology and Visual Science Research Center,
    Department of Ophthalmology, University of Pittsburgh School of Medicine, 203 Lothrop Street, Suite 835, Pittsburgh, PA 15213;
    wollsteing{at}upmc.edu.

Article source: http://www.iovs.org/content/50/3/1344.full

پاسخ دهید

نشانی ایمیل شما منتشر نخواهد شد.

*

code