Martin Brady^{1}, Raghu Raghavan^{1}, Andrew L Alexander^{2,3}, and Walter F Block^{2,3,4}

The hetergeneity of the brain makes designing a desired end drug distribution through pressurized catheters difficult. We present a method to utilize real-time MR monitoring of a co-infused Gd tracer during initial stages of the infusion to derive a real-time 3D estimate of the velocity front. We also describe a new algorithm that uses the velocity front to provide surgical feedback on the likely final infusion distribution.

Experimental delivery of convection-enhanced infusions in brain caner trials, depicted in Figure 2, begins with pre-surgical MRI (Fig. 2 Step a), then pre-surgical planning (Step b), followed by stereotactic OR guidance of a catheter inserted through a small craniotomy (Step c). Infusions begin in the OR on day 0 and often continue for 48 hours or longer. Post-infusion validation scans often show marked differences[1] in the actual drug distribution from the planned distribution, a shortcoming that is addressed here with real-time MR monitoring and periodic infusion prediction. We have previously shown how catheter alignment can be guided at 5 frames/second in the MR suite (Step c)[2], a rate approaching stereotactic OR feedback. The proposed method would continuously monitor a co-infused MR Gd tracer during the first few minutes to hours of the infusion protocol (Steps d-f in Fig. 2) to produce more accurate predictions of the actual end distribution.

To develop a new prediction algorithm utilizing real-time MR monitoring of the infusion front to infer the likely final distribution, we utilized data acquired in CED infusions in the putamen of a preclinical model. Volumetric T1-weighed acquisitions, altering between flip angles of 6 and 34 degrees, were acquired concurrently with the ongoing infusion (100 µl with 2mM gadoteridol infused at 1 µl/minute over 100 minutes). The acquisition pairs were used to obtain a time series of quantitative T1 maps, and the change in T1 provided a time-resolved concentration map of the ongoing infusion. A 3D velocity boundary vector at the infusion front was derived from a difference image between sequential volumetric monitoring maps.

The infusion flux at the infusion front is directly
proportional to the pressure gradient along the boundary according to dp/dn =v_{n }Φ/K where p is pressure, v_{n} is the
MR-derived velocity boundary vector, Φ is the pore fraction provided through a
previously acquired ADC map, and K is the hydraulic conductivity. By calculating the infusion flux at a
moderate distance from the catheter tip at an intermediate time point (tens of
minutes after infusion), the remainder of the likely infusion distribution can
be calculated using knowledge of the properties of the surrounding untreated
tissue, derived from pre-infusion MR imaging.

Pre-operative MRI is used to segment the brain as an input to a
commercial, brain drug transport algorithm (iPlanFlow^{TM}, Brainlab)
that provides predicted drug distribution for given catheter(s) placement, flow
rate, and flow duration.

Infusion workflow moves from pre-operative MRI
(Step a) to surgical planning( Step b here, also depicted in Figure 1). Catheter guidance is typically performed
using stereotactic guidance in OR with pre-surgical MRI, though real-time MR
can provide guidance into the actual surgical field (Step c). The proposed
methodology would monitor the infusion distribution for the first tens of
minutes to hours in a MR suite (Step e).
This work describes how a MR-derived velocity boundary vector can
provide an input to re-predict the end drug distribution (Step f).

Estimated end infusion distribution from commercial
pre-surgical planning algorithm predicts a largely spherical distribution
throughout six continuous slices of the putamen. Color-mapped concentration in
the range of 0 – 1.0 mM (0 – 50% of the infused concentration) is overlaid.

Estimated final concentration for
a 100-minute infusion at 1 μl/min, starting from a measured velocity boundary vector
at 22 minutes shows a more ellipsoidal distribution than shown in the
presurgical estimation in Figue 3. (6
axial slices shown, 0.8 mm separation.) Color-mapped concentration in the range of 0 – 1.0
mM (0 – 50% of the infused concentration) is overlaid. The
final concentration is not calculated within the interior of the initial
infusion (black region) front that initializes the algorithm.

Actual measured distribution at
the end of the 100 minute infusion.
Ellipsoidal boundaries show strong agreement with the infusion
distribution estimate derived from MR real-time monitoring shown earlier in
Figure 4.