| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Developments |
1 From the Dept of Electrical Engineering, National Taiwan Univ, Taipei, Taiwan, ROC (T.Y.H., H.W.C., M.Y.C., Y.J.L.); Dept of Radiology, Tri-Service General Hosp and National Defense Medical Center, 325, Sec 2, Cheng-Kung Road, Neihu 114, Taipei, Taiwan, ROC (T.Y.H., H.W.C., L.H.G., S.C.C., C.S.L., C.Y.C.); and Depts of Biomedical Engineering (M.Y.C.) and Radiological Technology (Y.J.L.), Yuan-Pei Univ of Science and Technology, Hsin-Chu, Taiwan, ROC. Received Jun 5, 2003; revision requested Aug 15; revision received Jan 14, 2004; accepted Feb 17. Supported in part by Tri-Service General Hospital with grant TSGH-C90–52 and by National Science Council with grant NSC90–2314-B-002–256. Address correspondence to C.Y.C. (e-mail: sandy0928@seed.net.tw).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
© RSNA, 2004
Index terms: Cerebrospinal fluid, flow dynamics, 167.12144 • Cerebrospinal fluid, MR, 167.12144 • Magnetic resonance (MR), phase imaging, 163.12144
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
Measurements of CSF production rate have been performed by using ventricular catheterization (3), external ventricular drainage (4), or ventriculocisternal perfusion (5) techniques. All of the above approaches, however, are highly invasive. Velocity-sensitive magnetic resonance (MR) imaging involving phase-contrast techniques has recently evolved as an attractive alternative for quantitative assessment of the CSF production rate (6,7), primarily owing to the noninvasive nature of this modality. For instance, a previous investigation involving phase-contrast MR imaging revealed circadian variations in the CSF production rate in healthy adults, with the minimum production rate (12 mL/h ± 7 at about 6:00 PM) being approximately 30% of the maximum rate (42 mL/h ± 2 at 2:00 AM) (8). These results, along with experimental evidence provided by follow-up studies, including one that revealed that administration of the ß1-receptor antagonist atenolol inhibited CSF production (9), indicate that the phase-contrast MR imaging method seems to have increasing promise in the quantification of CSF production, in spite of being limited to use in obtaining supratentorial measurements.
It is notable, however, that the data reported in the literature for human CSF production rates vary to a great extent, especially for values estimated by using invasive methods (347–370 µL/min) (2,3,10) versus values estimated by using MR imaging–based approaches (420–700 µL/min) (7–9,11–13). As a particular example, an aging-related reduction in CSF production that was demonstrated by using the ventricular catheterization method (2) was not revealed in an MR imaging–based study (11), the results of which, on the contrary, showed that there was increased CSF production in aged subjects as compared with production in young adult subjects, suggesting that interindividual factors other than age may play a role.
Although it is possible that the discrepancies may have originated from the very different physical principles of the invasive approach and the MR imaging method, it is also possible that the technical difficulties inherent in the MR imaging method were potential sources of inaccuracies. Specifically, the derivation of the supratentorial CSF production rate in the small cerebral aqueduct of Sylvius, which involves subtraction of two relatively large CSF flow rates in the caudal and cranial directions, is believed to be prone to inaccuracy because of error accumulation. Therefore, an optimization procedure for the noninvasive MR imaging measurement of the CSF production rate is certainly highly desirable. Thus, the purpose of our study was to determine the CSF production rate in healthy young adults by using cine phase-contrast MR imaging with high temporal and spatial resolution.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
The entire study, including the MR imaging sequence used, was approved by the institutional review board of Tri-Service General Hospital and National Defense Medical Center. Written informed consent was also obtained from all subjects. Note that because individual variations in CSF production rate are relatively large (as will become clear in a later section of this report), we did not attempt to perform a power analysis for experiment design based on the sample size requirement after obtaining some preliminary data. Instead, we aimed to focus on the technical developments involved in achieving an objective and precise measurement method.
The 23 subjects were imaged at night, between 10:00 PM and midnight. The time period for MR imaging examinations was prearranged because the CSF production rate in humans was once reported to exhibit circadian variations (8). For the purpose of velocity mapping, a 6-mm section perpendicular to the distal third of the aqueduct of Sylvius was selected by referencing a midsagittal image, as shown in Figure 1. A two-dimensional cine phase-contrast gradient-echo sequence (45/12, 20° flip angle) was used to obtain phase images with bipolar velocity encoding gradients set at a maximum encoding velocity of 20 cm/sec. Sixty-four cardiac phases were acquired and rearranged with retrospective electrocardiographic gating (14) to form 30 images that represented sequential phases in a cardiac cycle.
|
Automatic Aqueduct Segmentation and Image Analysis
Derivation of CSF production rate was performed with a personal computer (M2NE; ASUS, Taipei, Taiwan) after the phase images were digitally transferred from the console of the MR imaging unit. To minimize the influence of gross brain motion and eddy currents on phase values, the velocity values were first corrected with respect to nearby static tissue (7). A rectangular region of interest of about 1 x 1 cm2 on the velocity map was then selected (by T.Y.H., who had 6 years of experience in brain MR imaging) to encompass the aqueduct, after which all the analysis steps were executed automatically. Automatic segmentation of the aqueduct was performed by using a method modified from a pulsatility-based segmentation approach (15). The segmentation approach was based on the assumption that the variations in temporal velocity of the aqueductal pixels within a cardiac cycle should strongly correlate with each other. Owing to the nature of the laminar spatial pattern in the absence of turbulent flow, the above assumption was anticipated to be valid.
Note that the absence of turbulent flow in the aqueduct of Sylvius can be shown by calculating the dimensionless Reynolds number (R) with the equation R = (2
av)/
, where is the material density of the CSF, is the viscosity of the CSF, a is the radius of the aqueduct, and v is the average flow velocity. Because CSF is a fluid that has density ( = 1.0 g/cm3) and viscosity ( = 0.01 poise) similar to those of plasma (16), the maximum Reynolds number of the CSF flow, at a peak velocity of 10 cm/sec or lower and in an aqueduct smaller than 3 mm in radius (note that the parameters used are all worst-case estimations), was around 600. Thus, the estimated maximum Reynolds number was much smaller than the threshold of 2000 for unstable hydrodynamics (17) and strongly suggested an absence of turbulent flow.
With this principle, the temporal velocity profile of the pixel showing the maximum CSF flow velocity was first derived; then, the Pearson correlation coefficients given the above reference velocity profile were calculated for the temporal velocity profiles of all the other pixels. In this manner, a correlation coefficient map, which exhibited better immunity to noise fluctuations than did the original phase-contrast velocity map, was created. The area of the aqueduct was subsequently defined from the correlation coefficient map by using a region-growing algorithm (18).
The spatially averaged velocity as a function of cardiac phase was computed by summing all velocity values over the area of the aqueduct and then dividing the summed value by the area for all 30 phase images. The temporal profile of this spatially averaged velocity value was used to investigate the undersampling error, as will be described later. The CSF production rate was calculated as the net CSF volume flowing through the aqueduct in the craniocaudal direction in one cardiac cycle, multiplied by the heart rate. This was achieved by means of numeric integration of the CSF flow velocity throughout the entire cardiac cycle, as well as across the aqueduct area. It will be noticed that because all of the measurements were performed in the aqueduct, the CSF production rate reported in this study reflects supratentorial CSF secretion from the lateral and third ventricles only.
Operator-dependent Imprecision and Undersampling Error
To further emphasize the importance of objective segmentation of the small aqueduct, we compared the CSF rate calculated with the method described above with that computed by using a manual definition of the aqueduct on velocity maps. For the latter definition, CSF production rates of the 23 subjects were estimated by using the procedure stated above, with the exception that the aqueduct region was outlined independently by each of three operators (T.Y.H., M.Y.C., and Y.J.L., all of whom had 
5 years of experience in brain MR imaging). The interoperator agreement was then assessed.
In addition to operator-dependent imprecision, the possible influence of the temporal resolution used at image acquisition on measurement accuracy was also investigated. This latter investigation was performed by using Fourier transformation of the temporal CSF velocity profile, which yielded the power spectrum of CSF velocity at different frequency components (7) (up to the 15th harmonic in this study). A possible error caused by sampling the CSF velocity waveform with insufficient temporal resolution was estimated by examining the magnitude of the high-frequency components relative to the zero frequency component, a process that results in a value that is simply the net CSF flow velocity.
Data and Statistical Analysis
The interoperator agreement for CSF production rate measurements obtained by using manual aqueduct segmentation was assessed by reporting the range of estimation values in units of percentage of the mean of three measurements (ie, the best estimate when the true value is unknown). The average imprecision was defined as the absolute percentage difference between the individual measurements and the mean value, averaged for all 23 subjects. In addition, Kendall coefficients of concordance were calculated for the three operators to assess the agreement in continuous scoring (19). The Kendall coefficients of concordance ranged from 0.00 to 1.00, with 1.00 representing excellent agreement among the raters. On the other hand, the mean and standard deviation of the CSF production rates obtained between 10:00 PM and midnight in the 23 healthy young adults were also reported according to data derived by using the automatic segmentation algorithm.
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
|
|
|
Figure 4a, in which positive velocity represents CSF flow in the craniocaudal direction, shows the temporal profile of the spatially averaged CSF flow velocity in one subject; Figure 4b shows the corresponding Fourier coefficients of all of the different frequency components in the form of a power spectrum. The characteristic biphasic (ie, to-and-fro) waveform of the temporal profile (20,21) is clearly discernible in Figure 4a. The biphasic waveform is also appreciable in the power spectrum of Figure 4b, which shows that the fundamental frequency component (ie, the first harmonic, which corresponds to a heart rate of 60 beats per minute in this case) has a relatively large magnitude compared with the magnitude of the zero-frequency component (ie, the net CSF flow velocity). In fact, for all of the subjects recruited in this study, the sum of the first three harmonics in the power spectra—once described as constituting the essential waveform of the CSF flow profile (7)—was larger than the zero-frequency component (net CSF flow) by a factor of about seven to 11.
|
|
The mean CSF production rate measured in the 23 subjects by using the automatic analysis algorithm was 305 µL/min ± 145 within the time period from 10:00 PM to midnight. The estimated values were in good agreement with the reported CSF production rates of 347–370 µL/min obtained by using ventriculolumbar perfusion measurements (10) but were lower than those (420–700 µL/min) obtained by using early MR imaging methods (7–9,11–13).
|
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
It should be understood, however, that measurement of CSF production rate from the aqueduct of Sylvius has some intrinsic difficulties. The CSF in the cerebral aqueduct is characterized by a very small net degree of movement superimposed on a large-amplitude biphasic flow driven by the arterial pulse (20,21). Consequently, most estimation errors, in our opinion, are likely to originate from the accumulation of imprecision (ie, propagation of error) when one is deriving a small quantity from the subtraction of two large numbers. To resolve the discrepancy between MR imaging–derived CSF production rates and literature values obtained by using invasive approaches, it is therefore ultimately important to analyze the imprecision of the technique itself before attempting to attribute the discrepancies as arising from, for example, physiologic variations.
In the present study, variations in measured CSF production rate caused solely by manual definition of the aqueduct area were as large as 23.1% on average and ranged from –59.1% to +66.6%. Results of previous investigations (7,14) had indicated that the calculation of flow rate (or "stroke volume") is less prone to errors resulting from manual selection of regions of interest than is the calculation of average flow velocity, owing to the mutual cancellation of errors in area estimation and errors in average velocity estimation that occurs when one is attempting to derive flow rate. The results of our study, however, suggest that this is not the case for measurements of the "net" CSF production rate in the aqueduct of Sylvius.
Because the net CSF flow through the plane used at MR imaging of the aqueduct of Sylvius is inherently very small, we anticipated that an inclusion of nonmoving background tissue elements in the imaging plane may result in the introduction of relatively large errors through a magnification-of-noise effect. This presence of substantial imprecision in the subjective measurement of CSF production rate has at least two important indications. First, the spatial resolution must be sufficiently high to allow a clear delineation of the cerebral aqueduct, which is only about 2–3 mm in diameter. Second, dedicated automatic software for reliable and objective aqueduct definition that is immune to image noise is a necessary prerequisite.
If either of the above criteria were not met, measurement inaccuracy could have led to misleading results. It is noted that two of the operators in our study (M.Y.C. and Y.J.L., each of whom had 5 years of experience in brain MR imaging) were not informed beforehand of our purpose of interoperator dependency assessment, except for being asked to "try their best" in making reliable measurements. Therefore—additionally given the fact that a high in-plane resolution of 0.39 mm was used in our study—we regard the large operator variation reported in our study as close to the best-case scenario. On the other hand, because the image analysis algorithm employed in this study was based on pixel-by-pixel examination of the intrinsic similarity of flow patterns, it is therefore operator independent and yields high precision in a reliable manner.
The temporal resolution chosen for MR image acquisition is another important issue influencing accuracy in the measurement of CSF production rate. To understand how temporal resolution influences the accuracy of this measurement, one must realize first that the CSF production rate is determined by the net CSF flow velocity—that is, the zero-frequency component in the temporal velocity profile. The Nyquist sampling theorem states that discrete sampling of a periodic continuous waveform at a frequency of fs essentially results in a repetitive overlay of the spectra spaced fs apart in the frequency domain (22).
In other words, the use of 30 frames within one cardiac cycle in our study allowed the zero-frequency component to be overlapped only by the 30th harmonic of the temporal CSF velocity profile. Figure 4 shows that the magnitudes of the high-frequency components up to the 15th harmonic are at least comparable to the magnitude of the zero-frequency component. Consequently, errors due to undersampling aliasing could result if one uses a temporal resolution lower than about 8 frames within a cardiac cycle. These errors can be prominent, considering once again that the CSF production rate is inherently measured within a very small scale.
The CSF production rates measured in healthy young adults in our study were slightly less than, but in good agreement with, the reported value of 347–370 µL/min for ventriculolumbar perfusion measurements (10). This slight discrepancy is within reasonable expectations in that our measurement of the net CSF outflow through the aqueduct of Sylvius only registered CSF secretion in the lateral and third ventricles, while the CSF produced by the choroid plexus in the fourth ventricle may not have been taken into account.
Previous investigators who have evaluated phase-contrast MR techniques have reported values much larger than ours (8,9,11–13). We believe that these previous investigations may have been prone to resolution imprecision and operator-related inaccuracy, whereas we achieved high temporal and spatial resolution and used sophisticated image analysis software. Because the percentage of imprecision was not detailed in those reports (8,9,11–13), it is not clear whether those investigators’ reported values are significantly larger than our estimations in a statistically meaningful sense. In our study, operator-related imprecision alone ranged from –59.1% to +66.6%, even at MR imaging performed with 30 cardiac phases and 0.39-mm in-plane resolution. Hence, we anticipate that the actual imprecision in investigations involving the use of, for example, six cardiac phases should be even worse. Further studies performed by other independent investigators are needed to resolve the disagreement among MR imaging–based measurements of the CSF production rate.
There were at least two major limitations in our investigation that should be mentioned. First, the CSF production rate measurements were confined to the supratentorial region; hence, the CSF secretion from, for example, the choroid plexus in the fourth ventricle was excluded. Because the relative role of the fourth ventricle, as compared with the role of the lateral and third ventricles, in CSF secretion is largely unknown, our neglect of this portion might have had some influence on the interpretation of the results. We chose to examine supratentorial CSF production simply because the aqueduct of Sylvius, which shows largely biphasic laminar flow, should have served as a convenient target for reliable imaging measurements. Second, the large intersubject variations in CSF production rate among our subjects illustrate the difficulty in the potential usage of CSF production rate as a clinical diagnostic index. This difficulty is understandable, because any disease-related changes in CSF production rate must be large enough to exceed the normal intersubject variations to be statistically meaningful. To resolve this issue, further studies involving patients with CSF-related diseases are certainly needed.
In conclusion, for quantitative studies of the physiologic and pathologic mechanisms of human CSF secretion, it is important to establish a reliable procedure for measuring the CSF production rate. Our results show that two-dimensional cine phase-contrast MR imaging can be used for estimating the CSF production rate—if objective analysis can be performed with images obtained with sufficiently high temporal and spatial resolution. By using our approach, we obtained CSF production rates that are in good agreement with literature values that were obtained by using invasive ventriculolumbar perfusion measurements. Phase-contrast MR imaging is an effective technique that can potentially help resolve unanswered questions relating to human CSF secretion as well as to the relationship between human CSF secretion and intracranial pathophysiologic features.
|
FOOTNOTES |
|---|
3 Current address: Dept of Radiology, Tri-Service General Hosp and National Defense Medical Center, Taipei, Taiwan, ROC.
4 Current address: Dept of Automatic Control Engineering, Feng Chia Univ, Taichung, Taiwan, ROC.
Abbreviation: CSF = cerebrospinal fluid
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, C.Y.C.; study concepts, C.Y.C., H.W.C.; study design, T.Y.H., H.W.C.; literature research, T.Y.H., M.Y.C.; clinical studies, T.Y.H., S.C.C., C.S.L.; experimental studies, T.Y.H., Y.J.L.; data acquisition, T.Y.H., M.Y.C.; data analysis/interpretation, T.Y.H., H.W.C., C.Y.C., L.H.G.; statistical analysis, T.Y.H., H.W.C.; manuscript preparation, T.Y.H., H.W.C.; manuscript definition of intellectual content, T.Y.H., H.W.C., C.Y.C.; manuscript editing, H.W.C., C.Y.C.; manuscript revision/review, H.W.C., C.Y.C., C.S.L.; manuscript final version approval, C.Y.C.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and Methods ResultsDiscussionREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. L. Estrera, R. Sheinbaum, C. C. Miller, A. Azizzadeh, J.-C. Walkes, T.-Y. Lee, L. Kaiser, and H. J. Safi Cerebrospinal fluid drainage during thoracic aortic repair: safety and current management. Ann. Thorac. Surg., July 1, 2009; 88(1): 9 - 15. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Burbank History of Uterine Artery Occlusion and Subsequent Pregnancy Am. J. Roentgenol., June 1, 2009; 192(6): 1593 - 1600. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. M. Zimmerman and K. M. Kebaish Intracranial Hemorrhage Following Incidental Durotomy During Spinal Surgery. A Report of Four Patients J. Bone Joint Surg. Am., October 1, 2007; 89(10): 2275 - 2279. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
= experimental data, solid line = data after fitting with the first three harmonic components, R-wave = R wave of cardiac cycle. Positive velocity values represent CSF flow in craniocaudal direction. Note that CSF flow is largely biphasic (ie, to and fro), with an average velocity close to zero, suggesting that the derivation of CSF production rate can be prone to inaccuracies because of error accumulation. (b) Fourier coefficients of each harmonic shown as a power spectrum indicate that the zero-frequency component (ie, the net CSF flow) is relatively small compared with the first three harmonics. Also note that owing to the small magnitude of the zero-frequency component, the higher frequency components are not negligible in terms of CSF production rate measurements. N's frequency component = frequency component number.