Ocular MRI Imaging
The present invention relates to a method and apparatus for imaging the human eye or a portion thereof. More particularly, the invention relates to a method and apparatus able to produce a three dimensional image of the eye or a portion thereof.
Certain human medical conditions result in, or are a result of, anomalous ocular growth. One of the most notable conditions that fall into this category is myopia. Myopia is due to excessive growth of the posterior segment of the eye - in essence the eye outgrows the focusing power provided by the cornea and internal crystalline lens. In global terms, the increased prevalence of myopia in young adolescent children has reached epidemic levels - in some East Asian societies prevalence levels of 80% are not uncommon. Prevalence in western industrialised societies has reached at least 25% and is rising. Obtaining an understanding of the nature of ocular growth is central to an understanding of myopia and other conditions.
Very recent work indicates that the three dimensional (3-D) modelling of the eye may hold important clues as to how eye growth in myopia becomes abnormal. This work has also shown how calculation of ocular volumes can identify eyes that are susceptible to uncontrolled growth. These eyes can exhibit a posterior staphyloma that requires surgical intervention.
Advanced techniques for the measurement of intraocular blood flow have been applied to ocular pathology associated with glaucoma, diabetes and age-related maculopathy. Eye volume is an important variable in relation to choroidal blood flow (which accounts for 85% of intraocular blood flow). Interpretation of these measures would be facilitated by accurate determination of eye volume rather than using the relatively coarse approximations in current instrumentation.
3-D modelling of the eye could also assist in the design of image shells generated by a variety of ophthalmic appliances ranging from spectacle lenses, contact lenses, intraocular lenses and corneal inlays. Of special interest would be the potential to couple 3-D modelling of posterior eye shape with profiles of wave front aberration (expressed as Zernike coefficients) and to consider whether this would have utility in wave front guided laser corneal surgery.
Wave front aberration of the human eye is principally a consequence of the interaction between the quality of corneal optics, crystalline lens optics and posterior eye shape. As new surgical techniques develop (e.g. clear lens extraction as a method of treating presbyopia and myopia) there will be a need to understand the relative contribution of cornea and posterior eye shape to the formation of the retinal image.
Ocular Magnetic Resonance Imaging (MRI) has been previously used for 2-dimensional depiction of the human eye to investigate, for example, changes in the diameter of the ciliary smooth muscle collar with accommodation. Studies have used this information to comment on mechanisms of presbyopia (i.e. the ageing process that causes a need for reading glasses at around 45 years of age). Others have used 2-D MRI depiction to comment on posterior segment contour of the eye with a very recent paper using 2-D MRI representation to investigate separately posterior retinal shape in horizontal and vertical meridians. This work is described in the following:
Atchison DA, Jones CE, Schmid KL, Pritchard N, Pope JM, Strugnell WE, Riley RA Eye shape in emmetropia and myopia Invest Ophthalmol
Vis Sci 2004; 45(10) 3380-3386
Miller JM, Wildsoet CF, Guan H, Limbo M, Demer JL. Refractive error and eye shape by MRI. Invest Ophthalmol Vis Sci 2004; 45: ETVbstract 2388.
Cheng H-M, Singh OS, Kwong K et al. Shape of the myopic eye as seen with high-resolution magnetic resonance imaging. Optom Vis Sci 1992; 69: 698V01.
Ultrasound techniques have previously been used (e.g. B-scan) to generate a 2-D representation of the human eye, but owing to its relatively poor resolution and to the uncertainty in locating specific sector orientations, the precision for accurate shape representation is limited.
In terms of the 3-D representation of the eye, a computational method has been reported [Logan, Gilmartin and Cox (2002) presented as a poster at the ARVO Conference in the USA] which made mathematical estimates of eye volume based on rotation of a 2-D eye section synthesised from measurements of posterior retinal contour, components of axial length (using ultrasound), corneal topography and standardised measures of equatorial eye shape based on previously published MRI pictures. The technique does not represent the eye in 3-D as such, but rather generates a volume from 2-D data. Such construction of a schematic eye from various separate measurement parameters is limited in value as the parameters employed each have their own measurement error. Furthermore, inferring 3-D properties from 2-D information will always incur measurement error, especially in biological organs such as the eye, which are inherently spatially inhomogeneous. For example, there are substantial asymmetries within the eye, e.g. the optical and visual axes do not coincide. There is also a practical limit to the ocular information that current ophthalmic measurement systems can provide, as these cannot be invasive.
According to a first aspect of the present invention there is provided a method of imaging a human or animal eye, the method comprising: configuring a magnetic resonance imaging instrument to provide high contrast between fluid and non-fluid materials; acquiring a three-dimensional image of the eye and its surroundings using the magnetic resonance imaging instrument when so configured; and processing the acquired image to obtain a three-dimensional envelope of the eye or a portion thereof.
Embodiments of the present invention provide a means for generating a 3-D representation of the eye that is spatially manipulable. The invention allows calculation of eye shape and curvature at any selected point across the whole surface of the eye.
As well as providing accurate morphological information for research into conditions such as myopia, embodiments of the invention may provide results that can be utilised to assist surgeons to locate accurately an area of the eye requiring ocular surgery. Similar applications can be envisaged for assisting in retinal detachment surgery (myopic individuals are especially susceptible to retinal detachment) and to squint surgery whereby re-insertion of extraocular muscles could be mapped accurately to globe dimensions.
Said step of processing the acquired image may comprise identifying the outer boundaries of the fluid filled chambers of the eye. In this case, posteriorly the envelope corresponds approximately to the position of the retina that encompasses the posterior vitreous chamber and anteriorly to the surface of the corneal endothelium that encompasses the anterior chamber. However, the initial envelope may be further modified so that it corresponds to another boundary separating different tissue types, for example the outer surface of the sclera.
In a preferred embodiment of the invention, the magnetic resonance imaging instrument is configured to provide a T2 weighted volume scan. The instrument may also be configured to provide a T1 -weighted scan, said step of processing the acquired image comprising incorporating features obtained from the T1 scan into the image obtained from the T2 scan.
The magnetic resonance imaging instrument is preferably configured by setting one or more of the acquisition parameters such that the system is differentially sensitive to fluid, rather than other tissue types. Normally this requires a T2-weighted image with long TR and TE, with typical parameters within the ranges shown:
TR >=1000 ms TE >=100 ms
Preferably, the step of processing the acquired image comprises adjusting the brightness and contrast of the acquired image data to distinguish the eye or portion of the eye from the surrounding regions. The adjusted image data may then be flood-filled. The step of generating the envelope may comprise identifying the flood-filled area of the image data, and applying a shrink-wrap algorithm. More particularly, the step comprises generating a sphere mesh around the identified area, centred on the geometrical centre of the eye. Using an iterative procedure, the vertices of the sphere mesh are shrunk until the vertices contact the surface of the identified area.
Preferably, a smoothing algorithm is applied to the shrunk mesh.
The method may comprise using a second imaging procedure to measure one or more dimensional properties of the eye, and using the measured property or properties to correct the acquired image before or during processing. For example, said second procedure may be used to measure an axial length of the eye, with the image of the eye obtained using the magnetic resonance imaging instrument being adjusted so that the axial length of the eye corresponds to the measured length.
The method may comprise processing the obtained envelope of the eye to assign colours to the surface of the envelope in dependence upon the properties of a surface region, e.g. distance from a certain point, radius of curvature, etc.
According to a second aspect of the present invention there is provided apparatus for imaging a human or animal eye, the apparatus comprising: a magnetic resonance imaging instrument configured to a provide high contrast between fluid and non-fluid materials; data processing means coupled to said magnetic resonance imaging instrument for receiving therefrom an acquired three-dimensional
image of the eye and its surroundings and arranged to process the acquired image to obtain a three-dimensional envelope of the eye or a portion of the eye.
According to a third aspect of the present invention there is provided a computer storage medium having recorded thereon a computer program for configuring a magnetic resonance imaging instrument to provide high contrast between fluid and non-fluid materials, for acquiring a three-dimensional image of the eye and its surroundings using the magnetic resonance imaging instrument when so configured, and for processing the acquired image to obtain a three-dimensional envelope of the eye or a portion thereof.
For a better understanding of the present invention and in order to show how the same may be carried into effect reference will now be made, by way of example, to the accompanying drawings, in which:
Figure 1A illustrates a transverse cross-section through a 3-D image of the front of a human head acquired using T2-weighted MRI configuration;
Figure 1 B illustrates a transverse cross-section through a 3-D image of the front of a human head acquired using a T1 -weighted MRI configuration;
Figure 2 illustrates a flood-fill processing step applied to the image of Figure
1 A;
Figure 3 illustrates a shrink wrapping mesh applied to an eye contained in the processed image of Figure 2; Figure 4 illustrates a smoothing process applied to the wrapped eye of Figure
3;
Figure 5 is a flow diagram illustrating key steps in the 3-D image acquisition and processing procedure;
Figure 6 illustrates two different views of a processed image of a normal eye, dressed to show distance of the surface from the corneal pole;
Figure 7 illustrates two different views of a processed image of a strongly myopic eye, dressed to show distance of the surface from the corneal pole;
Figure 8 illustrates two different views of a processed image of a normal eye, dressed to show radius of curvature of the surface;
Figure 9 illustrates a view of a processed image of a strongly myopic eye, dressed to show radius of curvature of the surface; Figure 10 illustrates schematically a lateral cross-section through the eye showing various axial dimensions that can be used to correct a scanned image; and
Figure 11 is a schematic block diagram of apparatus according to the invention.
Magnetic Resonance Imaging (MRI) is now a well-established technique for obtaining three-dimensional images of the human or animal body. Using MRI instruments, images with a resolution of a fraction of a millimetre can be obtained. A description of the theory underlying MRI and of MRI instruments is beyond the scope of this document. For this purpose reference might be made to: MRI from Picture to Proton. By Donald W. McRobbie, Elizabeth A. Moore, Martin J. Graves, Martin R. Prince. Publisher: Cambridge University Press (December 5, 2002). ISBN: 0521523192; MRI : Basic Principles and Applications. By Mark A. Brown, Richard C. Semelka. Publisher: Wiley-Liss; 3 edition (September 5, 2003). ISBN: 0471433101. What is of interest here, however, is the possibility to tune MRI instruments to generate images that exhibit maximum contrast (where the term "contrast" is used here in a general sense to indicate a measure of visual difference) between materials of different type, and in particular a recognition that by tuning an MRI instrument to provide maximum contrast between fluid and non-fluid areas, an image can be obtained in which the eyeball is very clearly delimited from its surroundings.
Figures 1A and 1 B show transverse cross-sections of 3-D images of the eye acquired using a Siemens Trio 3-Tesla MRI scanner, with scanning sequences that are optimised for revealing detail in the eye. Figure 1A was acquired using a T2 weighted Turbo-Spin-Echo sequence with the following parameters: 35-45 slices, 512x512 matrix, 256mm Field-Of-View, 1 mm slice thickness, TR=1240 ms, TE=124 ms, Flip angle=150 degrees, 6 averages, 4/8 partial-phase acquisition, GRAPPA acceleration factor of 4. Figure 1 B was acquired using a T1 -weighted spin-echo sequence with the following parameters: 14 slices, 384x384 matrix, 200mm Field-of-View, 1.5mm slice thickness, TR=458 ms, TE=8.7 ms, Flip angle=90 degrees, 2 averages.
The sequence described above for Figure 1A, is known as a HASTE sequence - Half Fourier Acquisition Single Shot Turbo Spin Echo. This technique is a heavily T2 weighted, high speed sequence using a partial Fourier technique, and has high sensitivity for fluid detection and an acquisition time which is faster than conventional spin-echo sequences. The rapidity of the acquisition is important because of the need to acquire a complete image of the eye in a short period of time, so that the patient is able to maintain a relatively stable ocular fixation, thus reducing motion-induced blurring. Figure 1A was acquired in 5.5 minutes. The enhanced fluid sensitivity is created by choosing relatively long TR and TE timing parameters. Because the eyes contain fluid-filled chambers, they are seen to be extremely bright, with high contrast relative to the surroundings, in the HASTE image. This means they can be automatically identified and extracted from the MR image using the procedures described below.
Both scans are spin-echo sequences (to reduce susceptibility distortions), and employ local shimming around the acquisition volume. Again, shimming locally around the eyes increases the spatial integrity of the images.
In the examples described here, the images were acquired using a whole-head radio frequency coil array, which allows parallel-acquisition techniques to be used (GRAPPA 4 in the image of Figure 1A). This again allows rapid acquisition of the images. However, it is anticipated that the best possible spatial resolution of the eye can be obtained, in a similar acquisition time, using a dedicated ocular imaging coil, consisting of a small single coil placed directly over the eye. Such coils are typically custom-made to order, and are not standard items supplied by the MRI manufacturer.
The image shown in Figure 1 B is a T1 -weighted scan which reveals the fine detail of the eye, particularly the lens and ciliary body. Whilst this image may be extremely useful for certain applications, it is not easy to generate from it accurate data relating to the envelope or tunic of the eyeball. The boundaries between sclera surface and the orbit are not as well defined as in the image of Figure 1A. Note that the two images shown in Figures 1A and 1 B are optimised for different purposes. It is proposed to use the T2-weighted images, as in 1A, for automatic segmentation and 3D characterisation of the shape of the eye. However, the fine soft-tissue anatomical detail revealed by T1 weighted images, such as 1 B, can also be used to provide other morphological information about the eye, such as the position and dimensions of the lens, configuration and dimensions of the ciliary body, the thickness of the sclera, and most importantly the position of the optic nerve and the fovea. These anatomical parameters can also be used to inform the visualisation of the 3D eye models described in this document. For example, the position of the Fovea, determined using the T1 weighted image shown in Figure 1 B, can be "painted on" to the visualisations shown in Figures 6 to 9 discussed below.
A procedure for processing an image acquired using the optimised T2 weighted scan, such as is illustrated in Figure 1A, will now be described. The result of this procedure is an envelope or anatomical tunic of the eye on which
measurements can be performed for the purpose of diagnosing abnormal conditions.
The 3-D MRI images collected by the scanner are loaded into a modified version of a program called "mriδdX". The basic program has been developed for analysing MR images of the brain, and is available for download from the web site of the Aston University, Birmingham, UK (www.aston.ac.uk). The program used here is modified to include some features for morphometric eye-shape analysis as will be described.
Using the mri3dX program, the brightness and contrast of the image are adjusted so that the eye takes on a uniform white appearance. This may be done manually, with an operator adjusting the viewed image to optimise these qualities, or may be performed automatically. The resulting image allows automatic segmentation (i.e. identification) of the eye from the rest of the image.
The operator then places the crosshairs in the centre of the bright eye and selects a flood-fill option, which causes those areas of the image having a brightness within some specified range (e.g. 25%) of the brightness of the location of the crosshairs, to be filled with a uniform colour. The entire globe of the eye, including anterior and posterior chambers, is then filled with this colour. This procedure is illustrated in Figure 2, which shows a 2-D section through the acquired 3-D image. This procedure may be fully automatic, making use of some pattern recognition software to locate the approximate centre of an eye, and performing the flood-fill based upon the brightness at this centre.
Once the software has labelled all voxels (that is the smallest distinguishable box-shaped part of the 3-D image), a surface mesh is fitted to the outside of the eye. Many algorithms are available for this purpose. A preferred algorithm is however a "shrink-wrap" algorithm, in which a triangular mesh model of a simple sphere is first constructed. [ see van Overveld, K., Wyvill., B. (2004).
"Shrinkwrap: An efficient adaptive algorithm for triangulating an iso-surface. The Visual Computer: International Journal of Computer Graphics. 20(6). 362 - 379.] This sphere consists of over 32000 vertices and is centred on the geometrical centre of the eye, such that the sphere encloses the whole of the shaded eye. An iterative algorithm then starts, which pulls the vertices of the sphere mesh inwards, toward the centre of the eye, until they "hit" a shaded voxel. The effect is to "shrink-wrap" the eye so that a triangulated mesh representation of the eye's surface is generated. This is illustrated in Figure 3. The advantage of this shrink-wrap approach is that each triangle of the eye-model maps directly back to a triangle on the sphere surface (i.e. the eye and the initial sphere are homeomorphic). This property is important when group statistical analysis of eye-shape across subject populations is to be performed, as it will be possible to spatially average parameters such as distance and curvature on a standard spherical object.
The constructed eye model clearly looks like an eye, but is jagged in nature. This is because the shrink-wrapping tightly wraps around the sharp edges of the voxels that make up the MRI image. These voxels might be, for example, 0.5x0.5x1 mm in dimension. This might appear to be a fundamental limit on the resolution of the approach described here, were it not for the fact that the eye's surface should be locally smooth. By smoothing the mesh (of Figure 3), e.g. taking a weighted sum of the local vertex positions at each point of the mesh, a greatly improved image can be obtained. The result is illustrated in Figure 4. Conceptually, this process can be thought of as using a mesh that has elasticity, so that it does not turn tightly round the edges of the voxels in the MRI image. It is thus possible to construct a representation of the eye that actually has an intrinsically higher-resolution than does the initial scan.
Figure 5 is a flow diagram illustrating the key steps in the 3-D image processing procedure.
Now that we have defined a 3-dimensional model of the eye, we can use it to calculate morphological parameters of interest and colour-code the model for
3-D visualisation. For visualisation purposes, a software package such as GEOMVIEW (a general mesh viewer) may be used to visualise the 'dressed' models. Visualisation allows the model to be rotated in real-time so that the eye can be viewed from any angle
Referring firstly to Figure 6, this shows a model, obtained using the above procedure, of an emmetrope (i.e. no refractive error). The image has been colour coded based on the distance of the surface from the corneal pole, where Dark green (5) < 20 mm, Lighter Green (4) 20-21 mm, Light Green (3) 21-22 mm, Blue 22-23 (2) mm, Magneta (1 ) 23-24 mm, Cyan 24-25 mm, Yellow 25-26 mm, Orange 26-27 mm, Dark Orange 27-28 mm, Red 28+ mm. For the eye shown, this is, as expected, standard. As well as showing absolute distance from the corneal pole, the contours also give the viewer a feeling of eye shape.
Figure 7 shows a model obtained for an eye of a person who is strongly myopic (-6.5D). The increased axial length can clearly be seen (the red region (6) represents > 28 mm). This is a known characteristic of myopia.
Another parameter of interest which can be used to colour-code the model is the local radius of curvature of the surface of the eye. Figure 8 shows the radius of curvature for an emmetropic eye (no refractive error) by way of a colour coding scheme. The blue region 10 identifies a region in which the eye is "pushed-in" compared to the mean radius of curvature (i.e. radius of curvature flatter than the mean), whilst red colours 11 are regions that are "bulging" out (i.e. radius of curvature steeper than the mean). The image shown in Figure 9 shows the same kind of curvature analysis, but this time for a -6.5D myopic eye. Images such as these can prove extremely valuable for determining how the shape of the eye is modified as myopia develops.
One potential disadvantage of MRI is that, even using spin-echo sequences and local shimming techniques, the image may be subject to spatial distortions that limit the accuracy of the 3D model. However, other, more accurate
imaging technologies can be used to correct MRI images. Although these techniques cannot themselves provide a complete 3D model of the eye, they allow some critical dimensions of the eye to be accurately measured (sub mm). Examples of some such dimensions are depicted in Figure 10. If these critical dimensions are known, the segmented MRI data can be post-hoc corrected using an affine spatial co-ordinate transform that best fits the MRI model to the measured lengths. Ocular dimensions are obtained for each eye, non-invasively, using one or more of the following:
1 ) The ORBSCAN (Bausch & Lomb) takes 40 slit-sections of the cornea during 2 scans, each scan lasting 0.7 seconds. Each slit section is similar to an optical section viewed through a slit lamp. The instrument thus provides a full detailed surface topography of corneal radius, corneal apex location and central and peripheral corneal thickness.
2) The IOLMASTER (Zeiss) uses partial coherent interferometry to provide high-resolution measures of overall axial length. Measurement of the anterior chamber depth is by image analysis of an Optical section'. In addition the instrument can measure corneal radius and visible iris diameter.
3) The ACMASTER (Zeiss) again uses partial coherent interferometry to provide high-resolution measures of lens thickness, corneal thickness and anterior chamber depth.
4) The Oculus PENTACAM system (Oculus, Giessen, Germany) uses a rotating Scheimpflug principle to provide measures of anterior chamber volume and height. Central and peripheral measures of corneal thickness and corneal volume are also available.
The correction to the MRI image may be applied at one of a number of stages in the processing of the image. For example, correction may be applied after filling of the image (Figure 2), or after shrink wrapping and smoothing of the image (Figure 4).
Referring to figure 11 , the apparatus 10 according to the invention can comprise an MRI system such as a Siemans Trio 3-Tesla MRI scanner, consisting of a 3 tesla magnet 12 with associated gradient coils, a transmit/receive body coil and a patient bed 16, coupled with a phased-array radio-frequency (RF) head coil 14 capable of collecting high resolution images of the head and eyes. The gradients are switched using a set of gradient amplifiers 18 and the coils are driven and monitored using RF amplifiers and receivers 20. All components of the scanner are controlled by the electronics and acquisition system 22. The scanner is controlled by a operator and data reconstructed using a host computer 24. For offline reconstruction of the eye models using custom algorithms described here, an image analysis computer is used 26. Accordingly, computers 24 and 26 can comprise a processor for data analysis and image processing.
The apparatus 10 is configured so as to allow the collection and analysis of high-resolution T1 and T2-weighted images of the eye in 3-dimensions.
It will be appreciated by the person of skill in the art that various modifications may be made to the above described embodiments without departing from the scope of the present invention.