|Home | About | Journals | Submit | Contact Us | Français|
To develop a protocol which optimizes contrast, resolution and scan time for 3D imaging of the human eye in vivo using a 7T scanner and custom RF coil.
Initial testing was conducted to reduce motion and susceptibility artifacts. Three-dimensional FFE and IR-TFE images were obtained with variable flip angles and TI times. T1 measurements were made and numerical simulations were performed to determine the ideal contrast of certain ocular structures. Studies were performed to optimize resolution and SNR with scan times from 20 seconds to 5 minutes.
Motion and susceptibility artifacts were reduced through careful subject preparation. T1 values of the ocular structures are in line with previous work at 1.5T. A voxel size of 0.15×0.25×1.0 mm3 was obtained with a scan time of about 35 s for both 3D FFE and IR-TFE sequences. Multiple images were registered in 3D to produce final SNRs over 40.
Optimization of pulse sequences and avoidance of susceptibility and motion artifacts led to high quality images with spatial resolution and SNR exceeding prior work. Ocular imaging at 7T with a dedicated coil improves the ability to make measurements of the fine structures of the eye.
Traditional techniques used in imaging the internal structures of the eye, such as Scheimpflug photography, ultrasound and optical coherence tomography, are limited by optical distortions, require contact with the eye, are unable to image through pigmented areas, or have limited penetration posterior to the iris and sclera. The ability to non-invasively image all of the internal structures without distortion makes MRI an exciting tool for ocular research. Historically, MRI yielded lower resolution images which made it challenging to accurately image structures of the eye which are often less than a few millimeters in size (1, 2). 1.5T MRI has been utilized to study ocular disease, anatomy and physiology, and refractive error, and in the past decade, the in-plane resolution has improved from around 0.4 mm to about 0.1 mm due in part to the use of small custom coils (3–9). Unfortunately, the improvements in in-plane resolution necessitated the use of multi-slice 2D Spin Echo sequences with slice thicknesses of 3 mm and scan times up to 5 minutes (1, 3, 9). These compromises cause loss of data from gaps in 2D slice acquisition, partial volume errors from large slice thicknesses, and motion artifacts from involuntary fixational eye movements such as microsaccades which occur several times per second. The combined effects of these can lead to significant errors when measuring the fine ocular structures (1, 9). High-resolution, 3D MR images of short duration are necessary to make precise measurements of the eye and examine subtle changes that occur with age or disease.
Higher field strength should allow an improvement in resolution and shorter scan times. MRI at 3T has been used in ocular imaging (10), but only for volume measurements of the globe and with methods that were not optimized to detect smaller ocular structures. 7T MRI has become an important tool for neurological research (11–13) but, to the best of our knowledge, has not been used to image the human eye. The gain in SNR makes ultra-high field MRI uniquely suited for imaging the structures of the human eye; however, it is not without its difficulties as artifacts also increase with increasing field strength (14). B1 inhomogeneity leads to spatially variable flip angles (15, 16), Bo inhomogeneity increases, and motion artifacts are exacerbated.
The purpose of this study was to utilize the SNR gain available with a 7T scanner and custom RF coil to establish 3D imaging protocols for use in ocular research. 3D spoiled gradient fast field echo (3D FFE) and inversion recovery turbo field echo (3D IR-TFE) sequences were compared and tested based on previous work demonstrating exceptional performance at 7T (13, 14, 17). Further testing was conducted in order to minimize artifacts, optimize contrast and resolution, and reduce scan time.
The Biomedical Sciences Institutional Review Board approved the study protocol. Subjects were educated on the purpose of the study and informed consent was obtained prior to enrollment. All subjects were screened using the MRI Safety and Screening Form before being imaged. Ten healthy subjects age 26 to 54 years were scanned on a Philips 7T Achieva system (Cleveland, Ohio) using a volume head coil (Nova Medical, Inc., Willmington, MA) for transmission and a custom single loop 4-cm RF coil (Rapid MR International, Columbus, OH) for reception.
The first step in optimization was to establish a subject preparation technique to reduce artifacts inherent to scanning the human eye in vivo. The initial subject preparation was based on the previous work of Bert and Patz (1, 2). The subject’s head was constrained with foam pads to reduce movement. The RF coil was placed over an eye, cushioned with gauze and held in place with a Velcro strap. The contralateral eye was used to fixate a small target which was taped to the inside of the transmit coil. The focusing and movement of the eyes is yoked in normally binocular subjects, thus this also controlled the direction and focusing of the imaged eye. Multiple techniques were explored to decrease artifacts including imaging with the subject’s eye open or taped closed, with different positions of gaze, and by switching the phase and frequency encode directions.
The next step in optimization was to measure T1 of the ocular structures. Three-dimensional FFE images were obtained with a TR/TE = 20/2.9 ms and flip angles from 5 to 45° using moderate resolution (FOV 65 mm, 260×260, 15 slices, 0.25×0.25×1.0 mm3 voxel, 1:12 min scan time). 3D IR-TFE images were obtained with a shot interval between inversion pulses TS = 5000 ms and TI from 450 to 3200 ms (TR/TE/flip angle = 4.8/1.9 ms/8°, TFE-factor 160, FOV 65 mm, 160×164, 16 slices, 0.4×0.41×1.6 mm3 voxel, NSA 2, 3:25 min scan time). Low resolution B1 field maps were acquired for flip angle correction (18). Only a small correction (about 20%) of the nominal flip angles was needed across the entire region since the FOV was small. ROIs were manually traced by three observers as in previous studies (1), and included the cornea (c), anterior chamber (ac), posterior chamber (pc), lens nucleus (ln), lens cortex (lc), ciliary body (cb), iris (i), and vitreous humor (vh). Figure 1 shows the position of these structures. The 3D-FFE ROI signal data were analyzed by regression of the linearized equation (19–21) using the corrected flip angles α.
For the IR-TFE data, theoretical signal was computed using Deichmann’s equations (22)
The total duration of the turbo-gradient echo readout is τ=TR·TFE-factor. For linear phase encoding, the inversion time is the time from pre-pulse to the center of k-space α-pulse (at ½ TFE-factor·TR=τ/2). The delay time after the last α-pulse to the next inversion pulse (TD = TS-TI-τ/2), and M0 is treated as a scaling factor. Using these equations, the signal was computed for a range of T1s using the T1 value from the 3D-FFE data as a starting value. For each tested T1-value, χ2 was computed and minimized to find the T1 and signal scaling factor generating the best fit to the experimental data. Signal scaling factors from the FFE and IR-TFE data fitting and a receive coil drop-off factor (fRCVR) with distance from the coil, were used to estimate the relative proton density/T2* scaling factor (CPD/T2* = PD·exp−TE/T2*).
The third step in optimization was to perform numerical simulations of the signal intensities of the ocular structures for FFE with various TR and flip angles, and for IR-TFE with variable TS and TI using the measured T1, CPD/T2* and fRCVR (Table 1). For FFE computations the standard signal equations were used (20), and for IR-TFE the formulas of Deichmann were used (22). Simulations were aimed at achieving the highest SNR and CNR in the shortest scan time.
Finally, further studies of healthy volunteers were acquired using parameters from the theoretical optimizations. Both longer scan times of 2 – 5 minutes and short scan times of 20 – 40 s with different resolution levels were explored. Multiple short scans were registered in 3D using FSL (FMRIB Software Library, University of Oxford), and added to generate the final image. SNRs were calculated from manually traced ROIs in both the individual and final summed images. CNRs were computed for the ciliary body and lens with the surrounding vitreous humor.
Unlike at lower field strengths, susceptibility artifacts were a significant problem at 7T. As demonstrated in Figure 2, field variability at the interface of air and tissue along the eyelid margin caused significant artifacts. For the open eye, artifacts were seen at both the upper and lower eyelids, with the typical pattern of signal misregistration leading to bright and dark regions and distortions spreading into the lens (Figure 2a). Taping the eyelid closed reduced the susceptibility artifact (Figure 2b). Imaging with longer TE increased the size of this artifact (not shown). Focusing the contralateral eye slightly upward rotated the globe such that the artifacts did not obscure the structures of interest (ciliary body and lens).
As established in previous work at lower field strength (1, 2), taping the imaged eyelid also reduced motion artifacts from blinking. Motion artifacts from normal fixational eye movements were also minimized when the phase encode was in the AP direction (Figures 2c and d) (2). Motion artifacts were more apparent with the FFE sequence than the IR-TFE sequences. Motion artifacts were further reduced with the use of scan times under a minute (23, 24). Figures 2e and f illustrate a single 2:25 min 3D-FFE scan as compared to four registered 0:30 s scans. As conducted at lower field strength, individual scans were acquired with a brief delay between scans to allow for relaxation of the eye (24). The individual scans were later registered and added for further analysis. The summing of multiple short duration scans of lower signal was also suggested during collaborative visits to The Ohio State University and Wright Center of Innovation by Dr. Samuel Patz on March 31, 2008 and Dr. Bruce Berkowitz on October 7, 2008.
T1 measurements from the FFE and IR-TFE data for the ocular structures are listed in Table 1. There are noticeable differences between the T1s from the FFE and IR-TFE sequences. Average values were used for signal optimization computations. Compared to published measurements at 1.5T, dense structures such as the lens nucleus and cortex have longer T1 at 7T, whereas fluid filled areas including the vitreous humor and anterior chamber have similar T1 values. Differences between 1.5T and 7T for small structures (posterior chamber and ciliary body) may be due to partial volume averaging in the published lower resolution 1.5 T data. The PD/T2* factors CPD/T2* for FFE and IR-FFE were independently derived for both sequences from the amplitude fitting parameter A = S*CPD/T2**fRCVR by using a constant scaling factor (S) for all ocular structures and the receive coil drop off factor (fRCVR) computed as indicated in Figure 1. The relative proton density/T2* scaling factor CPD/T2* was chosen to be 1.0 for vitreous humor and the receive coil drop off factor fRCVR was set to 1.0 for cornea. With these choices, the resulting scaling factors were 950 and 3980 for FFE and IR-TFE, and are accounted for by the different spatial resolutions.
The average values for T1, CPD/T2* and fRCVR, (Table 1), were used for the simulations. Due to the primary author’s interest in presbyopia, we chose to focus our computations on optimizing contrast between the vitreous humor and the lens and ciliary body. Initial studies suggested that, visually, the best SNR and contrast were achieved with T1-weighted images with low signal for the vitreous, anterior and posterior chamber, and high signal for the lens and ciliary body (Figures 3 and and44).
Optimization of the FFE sequence is shown in Figure 5. To obtain the graphs in Figure 5, the optimum value of the flip angle that maximizes the signal difference between the vitreous and ciliary body was computed as a function of TR. The example TR and flip angle pairs provided are 10 ms/7° and 20 ms/10 °(Figures 5a and b). Optimal TR/flip angle pairs for the vitreous and lens were 10 ms/10° and 20 ms/14°. Next, signal divided by was computed for the optimal TR/flip angle pairs (Figure 5c). Since the SNR is proportional to the square root of the number of excitations, the signal divided by represents the achievable SNR per scan time (e.g., if the TR is halved, 2 averages can be acquired in that time leading to a SNR increase). While signal significantly decreased with decreasing TR, SNR per scan time (i.e. signal divided by ) remained constant. This simulation was confirmed in a 3D-FFE study with 4 repeat acquisitions at TR = 10 ms, 2 repeat acquisitions at TR = 20 ms and one scan at TR = 40 ms, yielding SNRs of 24.0, 24.1 and 24.2 for the lens nucleus. Alternatively, FFE contrast could be optimized for apparent proton density contrast (i.e. high signal for vitreous and low signal for lens and ciliary body with TR/α = 10 ms/1°). Generally, T1 contrast is optimal for flip angles twice the Ernst angle, and proton density contrast is optimal for flip angles approximately one-third of the Ernst angle.
Contrast in the IR-TFE sequence could be optimized via multiple approaches. We confined our evaluations to the following conditions: 1) optimizing contrast for vitreous and the ciliary body and lens and 2) acquisition of 3D high resolution images in less than 30 – 40 s per scan. Therefore, the shot interval between 180° pulses had to be less than 3 s and a large number of gradient echo read-outs (i.e. a high TFE-factor) were required. With this in mind, we computed two possible scenarios. First we simulated the signal for a range of TS and TI for short TR and moderate TFE-factor (τ = TR*TFE = 4.8 ms*160 = 768 ms), limiting in-plane resolution to 0.4 mm (Figures 4 and and6).6). Second, we simulated signal for longer TR and larger TFE-factors (τ = TR*TFE = 9.5 ms*260 = 2470 ms), allowing 0.15×0.25 mm in-plane resolution (Figure 6). For the low resolution scenario, signal and contrast between the vitreous and lens/ciliary body increased with increasing TI for fixed TS (Figure 4), and was maximized for the longest possible inversion time (TI = TS − τ/2). Analogous behavior was found for the high resolution case (not shown). Conditions were also tested for a range of TS: 1) TI=TS/2, 2) TI = TS − 1000 ms and 3) TI = TS − τ/2. In all cases, signal and contrast increased with increasing TS (Figure 6). More interestingly, signal per scan time (signal/TS) slightly increased with decreasing TS so the IR-TFE sequence improved with shorter TS. For all three TI scenarios, TS had to be larger than 2600 ms due to required pulse sequence delay times. Thus, all three cases are essentially the same since the entire time interval TS between inversion pulses is filled with TFE readout sequences, and TI≈TS/2. The ideal case for the IR-TFE sequence is to use the shortest possible TS for a given τ = TRmin*TFE and the longest TI allowed by the pulse sequence.
Finally, comparing the simulations in Figures 5 and and6,6, and given minimal TR of at least 10 ms due to gradient strength limitations (0.15 mm in read-out direction), the FFE sequence achieves a signal of 2.2 and 0.7 for the ciliary body and vitreous, whereas the IR-FTE signal is slightly lower at 1.8 and 0.4.
Initial calculations with ideal contrast and SNR indicated that a voxel size of about 0.03 mm3 is the maximal achievable resolution at 7T with our custom RF coil. This could yield multiple isotropic and anisotropic voxel size options (e.g. 0.3×0.3×0.3, 0.2×0.2×0.75 or 0.15×0.25×1.0 mm3). We chose to evaluate a voxel size of 0.15×0.25×1.0 mm3.
3D-FFE images at a resolution of 0.15×0.25×1.0 mm3 were acquired with multiple repeats with TR = 10 ms and flip angles of 3, 5, 10, and 20° (FOV = 65×65×15 mm, 420×260×15, scan time 36 s per scan). As expected from the simulations, CNR and SNR were highest for TR/α = 10 ms/10° (Figure 7). For this sequence, the average SNR of the lens nucleus was 21.0 for individual scans and 73.5 for 12 registered and summed scans. In the summed images, the resulting contrast between ciliary body and vitreous contrast and between lens and vitreous were 42.8 and 23.5. 3D-IR-TFE studies at the maximal resolution were acquired at TS/TI = 2600/1300, TR/TE/α = 9.5/2.9 ms/8°, TFE-factor = 260, FOV 65×65×8 mm, 420×260×8, scan time 34 s per scan (Figure 8). The average SNR for the lens nucleus in an individual image was 10.7 and the registered summed image of 12 repeat scans was 42.7. CNR for Ciliary body/vitreous and lens/vitreous contrast in the summed images were 54.7 and 30.2.
The selected TR and TE were the minimal values achievable for a 0.15 mm resolution in the readout direction. The limits on the TR and TE were due to gradient strength, not SAR issues. The gradient limitations determined settings for all other parameters, thus, for the given resolution, scan times of at least 30 s were required. In all cases, the SNR in the summed images was within 10% of the expected value from the average individual images ( ). This, and the quality of the summed images, confirmed that image registration was successfully achieved. Inter-observer variability for the ROI analysis was between 15 – 20%. Variability for repeat studies of the same subject on different days was between 20 – 40%. Overall, the FFE had higher SNR, but lower CNR than the IR-TFE sequence.
To our knowledge, this work represents the first in vivo imaging of the human eye at 7T. Although a direct comparison to lower field strength cannot be made without allowing for similar optimization, this study demonstrates the high SNR and resolution obtainable at ultra-high field strengths with a custom RF coil.
During the development of our scanning protocol we noticed a significant increase in artifacts with ocular imaging at 7T compared to lower field strengths. We were able to minimize susceptibility artifacts by taping the eyelids closed and instructing the patient to fixate with the contralateral eye such that the imaged eye is further from the eyelid margin. Rather than increasing the scan time to improve the SNR, we minimized motion artifacts by registering and adding multiple 30 – 40 s scans.
Although there are some reports on enucleated eyes, only Patz et al (1) has reported T1 and T2 measurements of the structures of the human eye in vivo. Their measurements were made at 1.5T on four adult subjects. Our initial T1 results at 7T exhibit trends that are consistent with published studies of T1 field dependence (25). As expected, the T1 values at 7T were higher than 1.5T measurements for dense structures (lens cortex and nucleus). The vitreous humor and anterior chamber (aqueous humor) of the eye are over 98% water with large T1 that should not change with field strength. Our 7T T1 values for the ciliary body were smaller than the 1.5T value, and our values for the fluid filled posterior chamber are higher. This is likely due to partial volume averaging of tissue with surrounding fluid for these small structures in the previous 1.5T data (1). Furthermore, our 7T T1 values of the fluid chambers (especially for the FFE sequence) are higher than what would be expected.. A recent study examined cerebrospinal fluid, also over 98% water, at field strengths varying from 0.2 to 7 T, and found a field independent T1 of about 4300 ms (25). Our T1 values for the fluid filled chambers are likely too long, because neither the FFE nor the IR-TFE measurements were optimized for measuring very long T1. Principally, T1 measurements could be improved by using longer TR and larger flip angle with the FFE method and longer shot intervals TS with IR-TFE, however this would lead to undesirably long scan times. We did not measure T1 for the optic nerve, retina, and extra-ocular muscles because the 4-cm RF coil we used for these studies was specifically designed for anterior segment imaging, thus there was a large signal fall off in the posterior globe. Other investigators (1, 24, 26, 27) have used TMJ and 3-inch surface coils with 1.5T scanners to more accurately image and measure the deeper ocular structures.
The numerical simulations and volunteer studies demonstrate that a final SNR over 50 was able to be achieved with a 7T scanner and custom coil. The optimized contrast for FFE and IR-TFE between the vitreous and the ciliary body and the vitreous and lens were over 23 and 30. These structures were initially selected for contrast optimization due to an interest in studying presbyopia. The SNR and CNR measurements were consistent between observers and over time with any variability likely due to RF coil placement. To the best of our knowledge, previous papers in vision science journals do not report SNR and CNR for comparison (4, 5, 8–10, 28, 29). Our final scan sequences yield a voxel volume of 0.0375 mm3, half that of previous 2D studies at 1.5T (3–5, 9). More importantly our scan durations of 30 – 40 s for individual 3D scans allow for accurate ocular imaging that avoids the common problems of motion artifacts, slice gaps in data, and partial volume errors. The total scan time of approximately 6 – 8 minutes is still reasonable for large scale studies of the general population. Most publications in vision science journals also do not specify acquired voxel size, which is the true limiting factor for SNR and interpolated voxel dimensions (4, 8, 9, 28, 29). Reporting only interpolated voxel dimensions can overestimate the true accuracy of the images. We report acquired voxel sizes throughout; images were interpolated 2 – 3 fold, such that displayed interpolated voxel size is 0.1×0.1×0.5 mm3 (volume = 0.005 mm3) (Figures 7 and and8).8). The improvements in resolution and scan duration are critical for precise measurements of the small changes that occur with age and disease.
Although this study was optimized for research in presbyopia, there are many potential clinical applications of ultra-high resolution eye imaging including studying both pathological and normal alterations in tissues, changes in ocular dimensions with refractive error, and the diffusion pathways of the eye, to name a few (6, 23, 30–33).
There are several ways 7T eye imaging can be improved in future studies. First, our data show a 4-fold signal drop-off from the cornea to the center of the globe/vitreous (~20 mm) with the 4-cm coil used in our initial study. Significant improvements in posterior imaging could be achieved using larger coils and/or multiple coils. Acquisition speed could be further improved with parallel imaging. Also, our simulations show that resolution is not limited by SNR but by the currently available gradient strength. In lieu of altering the available hardware specifications, improvement may be achieved by further parameter adjustment. We selected a fairly conservative FOV of 65 mm to avoid wrap-around from vasculature posterior to the globe. Anti-aliasing may allow for a smaller FOV, thus allowing increased speed with smaller matrix sizes and shorter TR and TE. Finally, our numerical simulations suggested an alternative strategy for improving SNR and CNR with IR-TFE. We noted that highest SNR and CNR were achieved with long inversion times. Yet, with the current pulse program implementation using linear filling of k-space, TI is limited to TI = TS−τ/2. This suggests that a reverse k-space filling scheme placing the longest TI in the center of k-space may further improve SNR and CNR.
The authors acknowledge Samuel Patz, PhD at Brigham and Women’s Hospital, Harvard, and Bruce Berkowitz, PhD at Wayne State University School of Medicine for helpful discussions, Hendrik von Tengg-Kobligk, MD for his initial 8T ex-vivo work of the eye, and Karla Zadnik, OD, PhD and Mark Bullimore, MCOptom, PhD at The Ohio State University College of Optometry for their assistance with this research.
Grant Support: This work was supported by the Ohio Department of Development (AGMT TECH 03-051 to the Wright Center for Innovation in Biomedical Imaging), the National Institutes of Health T32-EY013359 and K23-EY019097 and the American Optometric Foundation Ezell Fellowship to KR, and an unrestricted grant from Bausch & Lomb to the Ohio State University College of optometry for MRI research.