Measuring CSF Shunt Flow with MRI Using Flow Enhancement of Signal Intensity (FENSI)

Abstract

Purpose:

To develop and validate a non-invasive imaging technique for accurately assessing very slow cerebrospinal fluid (CSF) flow within shunt tubes in pediatric patients with hydrocephalus, aiming to identify obstructions that might impede CSF drainage.

Theory and Methods:

A simulation of Shunt-FENSI signal is used to establish the relationship between signal change and flow rate. The quantification of FENSI data involves normalization, curve fitting, and calibration to match simulated data. Additionally, a phase sweep method is introduced to accommodate the impact of magnetic field inhomogeneity on the flow measurement. The method is tested in flow phantoms, healthy adults, ICU patients with extra ventricular drains (EVD), and shunt patients. EVDs enable shunt flow measurements to be acquired with a ground truth measure of CSF drainage.

Results:

The flow-rate-to-signal simulation establishes signal-flow relationships and takes into account the T₁ of draining fluid. The phase sweep method accurately accounts for phase accumulation due to frequency offsets at the shunt. Results in phantom and healthy human participants reveal reliable quantification of flow rates using controlled flows and agreement with the flow simulation. EVD patients display reliable measures of flow rates. Shunt patient results demonstrate feasibility of the method and consistent flow rates for functional shunts.

Conclusion:

The results demonstrate the technique’s applicability, accuracy, and potential for diagnosing and non-invasively monitoring hydrocephalus. Limitations of the current approach include a high sensitivity to motion and strict requirement of imaging slice prescription.

Introduction

Hydrocephalus occurs in about 75,000 patients annually in North America, resulting in a need to divert cerebrospinal fluid (CSF) flow (1–3). The excessive CSF is usually drained by ventriculoperitoneal (VP) shunting (4). In practice, CSF shunts can become easily blocked. Although the outcomes of shunting practice on pediatric patients have improved over the last 30 years (5), about half of CSF shunts in pediatric patients still fail within two years, and shunt failure is the number one overall cause of pediatric neurosurgical interventions (1,6,7). One study compared two shunt systems with the standard differential pressure valves to assess their durability. 344 hydrocephalic children were included, and within 4 years, 59% of the valves had either obstructed, over-drained or become infected (6). The symptoms of shunt blockage are usually well defined (severe headache, nausea vomiting, lethargy and bulging fontanel in newborns), but at times are less typical, making it difficult to tell whether the shunt is functioning properly. Therefore, there is a critical need for a quick, accurate, and non-invasive shunt flow assessment method to determine when further intervention is indicated.

In those patients who are having symptoms of possible shunt malfunction, imaging of the brain is performed to determine if the ventricular size has enlarged. CT scan of the brain remains the most common radiologic test to assess shunt function. However, rapid brain MRI is preferred, as CT scans expose the brain, especially in children, to potentially harmful radiation. But MRI is not readily available in many clinical locations. The anatomical CT or MRI scan can have some limitations. For example, in some patients with shunt blockage the ventricular system does not enlarge. Also, a baseline scan is needed to compare with the current scan, which may or may not be readily available. Several invasive methods can be used to try to determine if the shunt is functioning properly. The reservoir can be tapped, and the pressure measured to determine if there is increased intracranial pressure. However, if the proximal ventricular catheter is blocked then it is not possible to record a pressure. A radionuclide technique assessing the ventricular CSF flow using a tracer has been reported (15), which later developed into a qualitative method called radionuclide shuntography; it uses a gamma camera and tracer to evaluate obstruction of the distal catheter (16). However, radionuclide shuntography has shown a large range of false negative rates in different studies, incorrectly indicating that the shunt is functioning correctly (no malfunction) in up to 36% of cases (17,18). This technique also exposes the patient to ionizing radiation. More recently, a noninvasive, non-imaging, thermodilution method has been developed, but its integration and accuracy in the clinical workflow is still developing (8).

Therefore, a better non-invasive imaging assessment is critically needed for evaluating shunt function, free from ionizing radiation. A quantitative measurement of shunt flow would be beneficial in patients with mild symptoms and could be recorded for comparison when the patient is asymptomatic. Magnetic resonance imaging (MRI) is an ideal non-invasive choice due to its use of non-ionizing radiation (9). MRI is not without challenges in this population, some shunts are programmable and have a magnetic adjustment; therefore, the set pressure can be changed during MRI. These shunts have to be assessed after any MRI and possibly reprogrammed. But more recently it was reported that there were programmable CSF shunt valves available that could be safely used in 3T MR imaging (10). Additionally, non-programmable shunts with fixed valves are in wide-spread use and are MRI compatible (11).

Several qualitative and quantitative methods have been developed based on MRI. A fast-field-echo T₁-weighted sequence with partial flip angle has been shown to be very sensitive to shunt flow through signal suppression effects and the long T₁ of CSF (12). More quantitative approaches have used phase contrast MRI for differentiating the reportedly obstructed and functioning shunts (13). Phase contrast is used in many quantitative techniques for flow measurements (14,15), but there are challenging spatial resolution requirements that must be met to achieve quantitative accuracy with flow rate (velocity) measurements in the shunt: the diameters of shunt tubes are 1~2 mm, and ideally 3–4 pixels must be included across the lumen of the shunt tube in both directions to achieve an accurate measurement (16,17). Phase contrast flow measurements face challenges in accurately measuring shunt flow with the unavoidable partial-volume effects under limited resolution (18). With the use of specialized coils to provide highly localized scanning at the site of the shunt, there have been successful approaches that have achieved quantitative accuracy in particular geometries (19,20) or with special flow-encoding settings (21). A recent phase contrast method using quantitative flow encoding, with the VENC tuned to target very slow flow, had a feasible and promising quantitative phantom test result, but showed degradation in performance under in vivo conditions, resulting in only qualitative indications of flow in patient samples (22).

Quantitative Flow Enhancement of Signal Intensity (qFENSI) (23), a blood flow measurement technique to measure slow blood flow or flux in the microcirculation, was previously developed and tested for functional imaging of reactive hyperemia. The current study expands this technique to measure coherent slow flow in CSF shunts. Shunt-FENSI aims to non-invasively and quantitatively evaluate the very slow shunt flow and determine if obstruction or infection is limiting CSF drainage in shunts. The new technique also includes a phase sweep approach in the tagging module to optimize the fitting of FENSI signals to mitigate the impact of magnetic field inhomogeneity on the flow measurement and provide a means to assess motion contamination of the signal. To demonstrate the effectiveness of Shunt-FENSI in measuring slow flow in shunts, simulations, flow phantoms, and patient scans were examined, including patients with extraventricular drains, where a measurement of drainage flow is available.

Theory

In the process of Shunt-FENSI, the experimental measurement and the quantification process are two equally important parts of the process. We describe the basic pulse sequence, focusing on the differences from the original FENSI sequence. Then we will discuss the quantification of the flow rate from FENSI data, which includes a simulated curve to quantify expected signal enhancement from the measurement and a normalization of the signal difference between scans.

Shunt-FENSI pulse sequence

The original sequence for the quantitative flow enhanced signal intensity (qFENSI) (23) was designed to capture slow flow rates of the cerebral blood flow (CBF). qFENSI is a control-tag subtraction technique with repeated saturation that was tuned up to the flow rates expected in microcirculation. We have adjusted the technique so that it is now sensitive to the very slow motion of CSF that is expected in vivo in the shunt, which is approximately 20 ml/hr (0.3 ml/min) based on healthy adult and pediatric flow rates measured in shunts (24,25). FENSI provides directional sensitivity to flow, giving a vector quantity of flux, with the highest sensitivity to flows perpendicular to its thin tagging plane. For maximal sensitivity to shunt flow, the tagging plane, and hence the imaging slice, is manually aligned perpendicular to the flow direction for which a measurement is desired. Fig. 1 demonstrates this arrangement. At the center of the imaging slice lies the very thin saturation (tag) slice. While the fresh spins are flowing into the tagging slice, the tag plane keeps saturating the spins within its region with tagging pulses, and a tag is gradually built up, and Fig. 1A–I shows the condition after tagging. A set of post-tag saturation pulses saturate the spins within a post-tag thickness in order to remove effects of the slice profile of the tagging, including both the tagged volume and static tissue in the region. Fig. 1A–II shows the condition within the imaging plane right after these procedures and before the readout. The segment of flow that escapes from the post-tag saturation plane, but remains in the thicker imaging slice, will generate signal from the control minus tag subtraction, providing a measure of the volumetric flow or flux information. It is worth noting that if the tagged spins flow fast enough to leave the imaging slice, no signal will be generated from the exiting saturated spins. In this condition, the flow rate exceeds our measuring range. Shunt-FENSI will also have a lower limit to the flow that can be measured, when spins are not fast enough to escape the post-tag saturation plane. The minimum and maximum flow rates desired for the measurement are important design considerations for the sequence and determine the tagging rates and thicknesses in the imaging protocol.

Fig. 1:

Diagrams of the tagging mechanism in Shunt-FENSI for (A) tag and (B) control versions. For both circumstances, the conditions at two time points are displayed. I. After the first tagging pulse, the tagging pulse tags the segment of flow within the tag thickness (orange). II. After all tagging pulses are applied and the flow has been moving forward, the tagging of the flow is built up. The post-tag saturation pulses are applied to saturate all tagging within the post-tag thickness (dark blue), and only the segment of the tagging that is outside the post-tag thickness remains, which contains the flow information. The control version has no tagging pulses applied; thus, no CSF is expected to generate signal. Tag thickness: 0.3 mm; post-tag thickness: 1.0 mm; imaging slice thickness (or voxel length, not fully displayed in the figure): 20 mm.

A control image, shown in Fig. 1B, is also acquired, where no flow is tagged. The post-tag saturation is still applied to achieve equal tagging of static spins in the tag region between the two conditions. In both tag and control images, a wide imaging plane is acquired after the saturation module. The difference between control and tagged signal intensity defines our signal enhancement, the source of flow information. This difference comes from the displacement of the moving flow within the imaging plane for a known duration. The higher the flow rate is, the more suppression happens in the tag signal. For this work, we are using a single-shot echo-planar imaging (EPI) readout with a spin echo to capture the FENSI signal as quickly as possible and with high sensitivity, so that the small tag from the very slow flow is able to be measured with several repetitions in an imaging experiment. Fig. 2 visually describes the pulse sequence within one repetition.

Fig. 2:

The schematic diagram of Shunt-FENSI pulse sequence. Purple gradients are slice select (ss) gradients for the tag/control tagging plane in FENSI. The green gradients are spoilers. The yellow gradients are ss gradients for the three-90° post-tag saturation pulses. The imaging portion of the sequence is a spin-echo EPI acquisition; the blue gradients are the ss gradients for the acquisition.

Saturation is accomplished with a series of paired 45° RF pulses in both the tag and control sequences. The Shunt-FENSI sequence repeats the pulse pairs 120 times over 3 seconds of tagging. When the RF pulses are applied in phase (as a +45° and +45° pulse on the same axis) and followed by a spoiling gradient, it saturates a thin tagging plane. The tagging thickness is set to 0.3 mm to maximize sensitivity to small displacements of CSF in the shunt. A post-tagging saturation is applied that is broader and saturates all the spins within 1 mm surrounding the tagging slice to eliminate side lobe effects from the RF tagging pulses. We define the thickness of the post-tagging saturation to be a factor of about 3 times wider than the tagging pulses, as determined from experimental results, to saturate side-lobe effects of the tagging pulse.

Within one acquisition, we have multiple TRs collected of Shunt-FENSI, with two adjacent TR’s (tag and control versions) having only one difference, the phase of the second RF pulse relative to the first. The RF pulse pairs are both applied on the same axis, resulting in adding two +45° RF pulses for the tag version TRs, and they are applied in opposite directions on the same axis as a +45° and −45° pulse pair for the control version. This 180° phase shift on the second 45° RF pulse ensures an effective 0° excitation for the control pair. We have to use equal energy RF pulses for tag and control conditions to avoid magnetization transfer tagging effects in the quantification of flow.

The tag effectiveness from one scan to the next can vary depending on the magnetic field inhomogeneity at the shunt location. Magnetic field inhomogeneity will result in the tag pulses not being applied in-phase and likewise the control pulses, not being exactly out of phase. To address this, we perform a phase sweep across consecutive TR’s in our acquisition, keeping all other parameters the same (the 180-phase shift between the tag and control pulse is also maintained), we use a stepwise set of phase offsets between the RF pulse pairs (one offset per TR pair) to sweep through one period of phase (0° – 360°) across multiple pairs of tag/control TR’s. Instead of just taking a fixed number of averages with the Shunt-FENSI measurement to improve SNR, we are able to accommodate magnetic field inhomogeneity across the measurements.

The phase offsets between the RF pulses result in a sinusoidal signal across the acquisition in the subtracted signal of control minus tag. We fit our signal profile and record the amplitude of the fit of the sinusoid as the Shunt-FENSI signal intensity, recorded as % signal lost in tagging, which is related to the amount of flow in the tube. The phase sweep component is vital for the Shunt-FENSI technique in general, since our previously reported method (26) only made one measurement and used averaging, resulting in spatially variable sensitivity and errors of underestimation of flow rates.

In the current phase sweep method used with Shunt-FENSI, the first three images collected by each sequence are usually corrupted in signal intensity due to steady state effects from the long T₁ of CSF and thus removed from analysis. The following 20 tag/control pairs, or 10 pairs of control and tag images, cover the phase range from 0° to 324°, while the phase step is 36° for each adjacent pair. We chose to collect 10 TR pairs because it gave enough resolution for the curve fitting and also resulted in a moderate total time for one scan session, with a 5 s TR, 23 tag/control images (3 discarded images and 20 usable images) taking less than 2 min.

Simulation of FENSI signal

To quantify flow rate during Shunt-FENSI imaging, we used a simulation of Shunt-FENSI to delineate the relationship between signal change and flow rate, incorporating T₁ recovery of the tagged CSF spins. The simulation result was used to find predicted flow rate from measured signal enhancement in the FENSI data processing, i.e. the “FENSI signal” or % signal loss in the tag versus control signal. We assume flow to be plug flow going through shunt. We note that it is important to simulate the 45° RF pulses and the time spacing between them as the control RF pulse pair does produce net tagging that depends on flow (27). In addition, previous studies on the simulation of the shunt-FENSI technique also showed that the relationship between flow rate and FENSI signal are highly influenced by T₁ relaxation rate over long tagging times (28). Shunt-FENSI has a total tagging time of 3 seconds, long enough for T₁ relaxation for CSF (using 4.163s as T₁ (29)) to affect the curve simulation, causing the relationship between flow and signal to be nonlinear.

The tagging timing and dimensions of the Shunt-FENSI protocol was set up to target flows in the range of 0.1 to 0.4 ml/min, representing shunt flow from a partially blocked shunt up to a fully functioning shunt at full flow rate. The upper limit of flow rate measurement, 0.4 ml/min, is defined based on expected healthy flow rates through shunts, which are 20 ml/hr or 0.33 ml/min (24,25). We designed the acquisition to be sensitive to flow rates within our chosen range and they plateau as the flow rate reaches the upper end of the designed range as the tagged flow starts to exit the imaging plane.

Quantification of Flow Rate

Quantification of flow by FENSI involves scaling the amplitude of the sinusoidal FENSI signal fit across the phase sweep data (% signal lost in tagging) by a reference value (with no tagging) and then converting the measured, scaled difference signal to a flow rate by comparison to the simulated curve. First, a corresponding reference scan (with 0° tag and control flip angles) is acquired to place the FENSI difference signal into a percent difference (or normalized signal) compared to a voxel occupied fully by CSF, such as from a ventricle. This conversion factor is used to normalize the data from different scans to the same scale based on the CSF or water measure in the reference 0° flip angle scans.

We then use a separate scaling factor to scale the flow measurement back to the simulated data by using data from a phantom calibration experiment with known flows. This scaling factor is determined once per Shunt-FENSI parameter setting in a controlled phantom experiment and used for other phantom and human experiments with the same settings. In our fitting process, the scaling factor was found from a minimum mean squared error fit between phantom data across 10 sets of flow data and the simulated curve. This value of the scaling factor was recorded and was tested with an additional 5 sets of flow data.

Methods

The phase-sweep Shunt-FENSI sequence was tested in flow phantom, healthy adults, intensive care unit (ICU) patients with extra-ventricular drains (EVD), and shunt patients. All the data was acquired in accordance with an approved IRB protocol. Like shunt patients, ICU patients with EVDs have hydrocephalus and have a tube placed to temporarily drain fluid (4,13,30). Unlike shunt patients, EVD patients drain fluids to an external bag, enabling the assessment of flow rates. Fig. 3 illustrates the typical positioning in the target slices in the phantom, healthy adult, and child tests. We note that the shunt tubing is dark in the figure due to the small cross-sectional area of the tube (diameter 1.5 mm) in the midst of the large imaging voxels (3.44 mm), filling only 15% of the voxel’s cross-sectional area.

Fig. 3:

The typical positioning in the target slices of the phantom, healthy adult, and child patient tests. (A) 4 shunts inside the holder beneath the phantom; (B) 3 shunts inside the holder beneath the head of a healthy human; (C) 2 shunts (1 open 1 closed) inside the brain of a pediatric patient.

Fig. 4 displays the full experimental setup for the phantom and healthy human tests. We customized a special-designed 3D-printed shunt holder which holds 4 straight segments of the shunt in parallel. The 3D model is shown in the upper-left corner of Fig. 4. In phantom and healthy adult scans, we wrapped the shunt in this holder to keep the tubes parallel to each other and perpendicular to the imaging plane, placed the holder beneath the phantom / the head of the volunteer, and used an MRI-compatible syringe pump (Harvard Apparatus, Holliston, MA) to produce flow through the shunt. The phantom and healthy adult scans were reported in the form of curves of the controlled flow rate versus the FENSI signal, so that it can be compared and verified with the simulated curve, described in the previous section in Fig. 3. With 4 perpendicular intersections of the shunt tube per image, we analyzed 20 total shunt tube locations in 5 phantom scan sessions, and 16 shunt tube locations in 4 healthy adult scan sessions. We determined that it takes 20–30 seconds for the pump to change the flow rate from 0 to 0.4 ml/min in the shunt between scans at different flow rate settings. Therefore, during scans involving the syringe pump, each FENSI sequence started 30 seconds after the pump was set to any new flow rate.

Fig. 4:

The experimental setup of the phantom scans and healthy human scans.

EVD patient tests are one of the main steps of validation. The proximal end of the EVD is placed in the ventricle and drains CSF into a drainage bag. The valve of the EVD drainage can be set open or closed when scanning the patient. When the drainage valve was open, we were able to have a measurement of the flow rate inside the EVD tubing by recording the duration and the fluid accumulation in the drainage bag during the course of the experiment. Since the flow rate in EVD tubing is known but uncontrollable, this is an intermediate step between the shunt holder tests, with known and controllable flow rates, and the actual application on shunt patients, which is an unknown and uncontrollable flow rate. We scanned 5 EVD patients with our sequence, but only 3 out of 5 scanning results are able to provide reliable flow measures, because the others are corrupted by excessive respiratory motion in the critical care, sedated patients.

In the scans of shunt patients, the primary target of this research, we used the phase sweep Shunt-FENSI sequence to measure the shunt flows inside the brain. In this type of test, the flow rates are not controlled and are unknown. We had 6 healthy shunt patients (no flow deficits expected) scanned (5 adults and 1 child), and only 4 of the adult datasets are usable; the 2 other datasets suffered from motion sensitivity and were discarded. In the 4 usable datasets, we were able to interrogate 5 shunts inside the brain (one patient had the same shunt scanned at two different locations). For one patient we additionally collected data from the shunt outside the skull with Shunt-FENSI.

For the Shunt-FENSI acquisition, we use a TR/TE of 5 s/48 ms, 25 ms between repeated tagging pulse pairs and spoilers, with the total tag duration to be 3 s. The field of view is 220 * 220 mm, and the resolution is 64 * 64. For our scans, we did 23 measurements, which include three measurements which are discarded due to steady state effects and the remaining 10 pairs of control and tag measurements. Each sequence duration is 1 min 55 sec. The thicknesses for the tagging, post-tag saturation, and imaging slice are set to 0.3 mm, 1 mm, and 20 mm.

An RF pulse of 0.3 mm thick is challenging to achieve and the tag thickness impacts our quantification of Shunt-FENSI signal. We measured the slice profile on a phantom and found that the full width at half max (FWHM) of our 0.3 mm excitation pulse was 0.55 mm. We incorporated this pulse width into our simulations for more accurate quantification.

To test if the updated sequence with phase-sweep feature properly sweeps through the phase offsets, we compared our phase sweep results with mis-tuning the RF center frequency of the magnet, at 10 steps of 36° phase offsets (using the time between the RF pulses in a pair as our reference time) on the flow phantom. The phase sweep sequence was acquired with its built-in incremental phase offsets but at the tune-up center frequency for the MRI system. The frequency check for phase-sweep method made sure that our sequence was sweeping properly through a 360° phase.

Our scan protocol included a localizer followed by a 3D T2 TSE sequence to locate the shunt. To get the most accurate signal for quantification of shunt flow, Shunt-FENSI requires that the imaging plane be perpendicular to the shunt. Graphically selecting the desired perpendicular imaging slice from the 3D view of the TSE data is required to get the correct reference. A high resolution 2D single-slice TSE is acquired of the planned imaging plane to provide resolution sufficient to view the shunt.

A calibration run of Shunt-FENSI is acquired with a 0° flip angle tagging and control pulses, but otherwise the same timing as the flow measurement runs, over just a few TRs. The average of these images serves as the reference in image registration and as the water reference flow quantification, as explained previously.

In addition, a set of sequences for T₁ measurement was included since the curve of the anticipated percentage signal loss is highly influenced by T₁ value of the draining fluid. T₁ values of drainage fluid can be modified in EVD patients due to contamination of the CSF with other fluids.

Results

Signal-to-Flow-Rate Simulation

Regarding the signal simulation, Fig. 5 shows the simulated percent signal change across flow rates for various T₁ values, which implies that T₁ does have an effect on this relationship. As stated, we assume a 4.163s T₁ for CSF (29), and use the red curve in Fig. 5 as the simulated curve to find the measured flow rate from the Shunt-FENSI signal, as described above in flow quantification.

Fig. 5:

The simulated percent signal changes across flow rates for different T₁ values.

Phase-Sweep Frequency Check

We collected a dataset on a phantom comparing our phase sweep results and purposefully mis-tuning the RF center frequency of the magnet, demonstrated in Fig. 6. The two curves show similar behavior and are both sinusoidal-like.

Fig. 6:

A comparison of the raw signal difference (ΔM) between the phase-sweep FENSI sequence and the manual adjustment of measuring frequency.

Quantification of Flow Rate

The calibration between the quantifying simulation and phantom experiment data is shown in Fig. 7. A calibration scaling factor of 11.2 is identified, with minimum mean square error of the average phantom data towards the simulated curve.

Fig. 7:

The calibration result of the average phantom test data with the simulation used for quantification.

Phantom and Human Scans

Fig. 8 shows the typical structural scans from an imaging session on the shunt and in an EVD patient showing the FENSI slice prescription relative to the shunt tubing. A high-resolution (hi-res) 2D T2 TSE image of the slice is shown with the tubes identified in the red box along with a Shunt-FENSI reference image (0° flip angle for tagging pulses). The T2 TSE 2D sequence is a higher resolution of 640*640, enabling visualization of the small shunt tubes, while the FENSI images are 64*64 for the field of view of 22 cm.

Fig. 8:

The images of one typical phantom test (A-E) and one typical shunt patient test (F-J) at the selected imaging slices. The red regions show the locations of the shunts. (A, F) T2 TSE 2D high-resolution; (B, G) FENSI tag; (C, H) FENSI control; (D, I) FENSI difference (ΔM) at one particular time point tag/control pair; (E, J) FENSI flow (after phase-sweep fitting). Note that in (D), the rightmost shunt of four shunt sections in the image is invisible since the sinusoidal fitting for it peaks at a very different phase compared to the selected phase; the flow in all four shunt segments are visible in the same image only after applying the phase-sweep fitting.

The quality of calibration and reliability of the Shunt-FENSI flow measure can be seen in the phantom and healthy human tests in Fig. 9A. Both types of data were quantified with the quantification process using calibration data from a separate phantom run. There is good agreement between the FENSI measured flow and the actual flow rate (set on the flow pump). Increasing variances for both curves are also noticeable as the flow rate increases.

Fig. 9:

The Shunt-FENSI validation results. The results are collected from (A) the phantom, the healthy human volunteer tests, and (B) the patients with EVD tubings.

Fig. 9B shows the Shunt-FENSI measured flow rate versus the flow rate measured in the drainage bag for the patients with EVD tubing. The FENSI measured flow in EVD 1 (yellow) and EVD 3 (dark blue) agree with the amount collected in the drainage bag, however, there is a suspected scaling problem in EVD 2 (light blue), which overestimates the flow rate, which may be due to movement between the Shunt-FENSI acquisition and the reference scan. The data collected with the shunt tube closed do not produce a 0 estimate of flow rate, instead, noise in the measurement results in a minimum measure of approximately 0.1 ml/min. In general, we see clear separation of the measured flow rate between the closed and open states from all data.

The VP shunt patient test is the application test of the Shunt-FENSI technique. The results are shown in Fig. 10. Although for these data we don’t have an external measure of the flow rate to compare to, the patients have no clinical symptoms, and we assume that all of them have normal, functional shunt tubing. The Shunt-FENSI measured flow rate indicates that most of the measured flow rates (excluding Location 1 of VP 2) are in the range of normal flow, although slightly lower than expected. The disagreement of Location 1 for VP 2 may result from a bad selection of the imaging plane perpendicular to the tubing, or an unexpected movement between the 3D planning scan and the Shunt-FENSI scan that changed this perpendicular spatial relationship.

Figure 10:

The utility test from the shunt patient tests. The actual flow rates for these data are unknown. Except for one data with special notation, all data are collected inside the brain.

Discussion

We developed, validated, and applied a method to measure very slow flow in ventriculoperitoneal shunts. With Shunt-FENSI, we can non-invasively assess shunt flow using noninvasive MRI, which is applicable in pediatric populations. We revised a method initially developed for imaging slow flow in the microcirculation and adapted it to the slow flow in a shunt, where flow rates are expected to be about 0.3 ml/min on average. In this study, we developed and calibrated the technique on a shunt holder placed beneath phantoms and healthy subjects’ heads. We used further validation in EVD patients, where the average amount of fluid drained could be measured and compared to Shunt-FENSI-derived flow rates.

We set 0.4 ml/min as our upper limit for the flow measurement because the CSF flow through shunts has been measured to be about 20 ml/hr (0.33 ml/min) in normal shunt patients (24,25). We tuned the parameters in Shunt-FENSI, including the number of pulses, the spacing between tags, and tagging / post-tag thickness, to achieve an upper limit of 0.4 ml/min. The lower limit of flow is determined by the post-tag saturation pulses which eliminate very slow moving flow from the measurement along with noise in the measurement impacting the Shunt-FENSI signal fit with no flow. For the current state, our lower limit lies around 0.1 ml/min, as both plots of the EVD patients and shunt patients show. This lower limit of measurement is determined by the settings in our Shunt-FENSI protocol and can be adjusted. However, a level of 0.1 ml/min provides a cut-off between robust non-invasive measures of slow flow with Shunt-FENSI and the need for further clinical evaluation when flow lower than expected is observed.

Shunt-FENSI was evaluated for its ability to measure very slow flow in phantoms, healthy humans, EVD patients and shunt patients. In the phantom tests, the measured flow rates agree with the actual flow rates well, while the other tests with known flow rates, Shunt-FENSI measures tend to underestimate the flow value at higher flow rates. This may be due to physiological noise impacting the fits or slight motion impacting the perpendicular alignment required for full measurement of the flow rate. In our human subjects, we also found that, by being sensitive to slow flow in the shunts, we are also highly sensitive to motion corruption. In Fig. 9B, where we compare Shunt-FENSI measured flow rates with those measured from the EVD drainage bag, there is a wider discrepancy between the measures. This may be due to a few factors. First, the “ground truth” measure of flow was obtained by measuring the discharge into the drainage bag during the entire experimental condition, which included modifying the flow rate by adjusting bag height or turning off the drain, tabling the patient in, running anatomical scans, and then running the Shunt-FENSI scan. There may be mismatches caused by having dynamic changes in drainage over time to the bag during the experimental condition. In addition, EVD patients are critical care patients and tend to move during the scan, even those that were not conscious during the scan. The sedated patients demonstrated large breathing motions in these cases. We also experienced challenges in controlling the position of the drainage bag, whose height controls the flow rate through the EVD tubing. Getting proper placement and conditions for the drainage bag during the MRI scan is critical to obtain flow rates within our ideal range. This could also happen in our target population, where gravity effects from being supine for a scan could impact flow rates in shunts.

As can be seen in Fig. 8, as Shunt-FENSI uses a single-shot EPI readout, there can be geometric distortions and fat shift artifact. The scanner operator must pay attention to the location of the artifact relative to the shunt tube location in the image being used for flow measurement. The operator can shift the phase encode direction from AP to PA if the fat artifact hits the recording location, which can be observed in the reference 0° flip angle image taken before the full Shunt-FENSI run. In addition, the operator could also decide to scan the tube at a different location with better spacing between the shunt tube and the fat layer to avoid this artifact as the flow in the tube should be the same along its length.

In the shunt patients, we scanned one patient twice at two different locations inside the brain for the shunt tubing: Location 1 and Location 2 of VP 2. These two measures provide a low measure of flow but differ in their quantitative value. It is possible that the subject moved between the anatomical scan which was used to find the perpendicular slice position and the acquisition of the Shunt-FENSI sequence. We used the same high resolution 3D TSE scan to localize the tube for each acquisition and the subject may have moved slightly between the two runs.

One limitation of our technique is its high sensitivity to motion including respiratory motion and rotations of the head. In order to be sensitive to the small displacements of CSF during a scan, we are also sensitive to other bulk displacements. We have demonstrated that we can measure shunt flow in conditions with minimal human motion in actual patients, but 36% of our human data from the EVD and shunt patients was corrupted by motion. To provide a metric to assess if motion corruption occurred, we examine the residual of the fit after fitting the phase sweep sinusoidal pattern across the time series. The metric is calculated as the normalized root mean square (RMS) of the residual of fitting (of control minus tag data) normalized to the RMS of the original, unsubtracted signal. If this metric is above 0.05, then we discard the data as corrupted by motion. Additionally, CSF flow itself can be dependent on cerebral arterial pulsation and respiration, varying across the acquisition which can affect the fit. The impact of this should be examined in future work.

Conclusion:

A non-invasive and quantitative method is presented to measure the slow flow of CSF in shunts using a method called Shunt-FENSI. Although sensitivity to patient motion was seen in many of the experiments, reliable estimates of shunt flow could be made in healthy human phantom experiments and in extra-ventricular drain patients, situations in which a gold-standard flow measure is available. Further, Shunt-FENSI provided reasonable estimates of flow in healthy shunt patients. The technique could provide a non-invasive way to monitor shunt function in pediatric patients.

Data Availability Statement

Sample data from one typical phantom test and one deidentified shunt patient test (as shown in Fig. 8), with the corresponding analysis MATLAB code, can be found at: https://doi.org/10.13012/B2IDB-7252521_V1.