|Home | About | Journals | Submit | Contact Us | Français|
Convection-enhanced delivery (CED) is a novel drug delivery technique that uses positive infusion pressure to deliver therapeutic agents directly into the interstitial spaces of the brain. Despite the promise of CED, clinical trials have demonstrated that target-tissue anatomy and patient-specific physiology play a major role in drug distribution using this technique. In this study, we retrospectively tested the ability of a software algorithm using MR diffusion tensor imaging to predict patient-specific drug distributions by CED. A tumor-targeted cytotoxin, cintredekin besudotox (interleukin 13–PE38QQR), was coinfused with iodine 123–labeled human serum albumin (123I-HSA), in patients with recurrent malignant gliomas. The spatial distribution of 123I-HSA was then compared to a drug distribution simulation provided by the software algorithm. The algorithm had a high sensitivity (71.4%) and specificity (100%) for identifying the high proportion (7 of 14) of catheter trajectories that failed to deliver drug into the desired anatomical region ( p = 0.021). This usually occurred when catheter trajectories crossed deep sulci, resulting in leak of the infusate into the subarachnoid cerebrospinal fluid space. The mean concordance of the volume of distribution at the 50% isodose level between the actual 123I-HSA distribution and simulation was 65.75% (95% confidence interval [CI], 52.0%–79.5%), and the mean maximal in-plane deviation was less than 8.5 mm (95% CI, 4.0–13.0 mm). The use of this simulation algorithm was considered clinically useful in 84.6% of catheters. Routine use of this algorithm, and its further developments, should improve prospective selection of catheter trajectories, and thereby improve the efficacy of drugs delivered by this promising technique.
Malignant gliomas (MGs) remain rapidly and almost uniformly fatal. Systemic delivery of many potentially effective drugs to these and other intracerebral tumors is hampered by the restrictive blood-brain barrier and high intratumoral pressure.1–8 The innovative intracerebral drug infusion technique of convection-enhanced delivery (CED) uses a positive infusion pressure to deliver therapeutic molecules throughout the interstitial space of brain parenchyma, theoretically resulting in homogeneous distribution of macromolecular therapeutic constructs at clinically relevant volumes and concentrations.9–13 The tremendous potential of this simple approach has been clearly demonstrated in preclini-cal2,9,10,12–26 and clinical studies by us and others.27–37
In previous human studies, we have shown that CED could produce extensive and relatively homogeneous distribution of iodine 123-labeled human serum albumin (123I-HSA) in the brains of patients with MGs.34 Although these initial studies confirmed the promise of CED, they also demonstrated that spatial distributions could vary significantly from patient to patient. Furthermore, the actual geometry of the distribution in a given patient was not obviously predictable. As a result, infusions frequently failed to reach the intended regions of infiltrating tumor such that optimum drug delivery occurred in as few as 20% of patients.34 Clearly, this variability constrains the potential of this approach and ultimately the efficacy of the therapeutic agent being delivered.
Based on theoretical considerations and analysis of our preliminary images, our assumption was that inter-patient variability could be explained by disparities in the physiology and anatomy of different brain tissue regions. Although these disparities cannot be fully appreciated with conventional anatomical MR images, our mathematical models suggested that diffusion tensor imaging (DTI) could provide much of the necessary information. The pilot study reported here assessed the clinical usefulness of a computer algorithm based on these assumptions to predict distribution by CED of large molecules infused into the brain. Seven adult patients with recurrent MGs were administered cintredekin besudotox (IL13 [interleukin 13]-PE38QQR) along with 123I-HSA as a surrogate imaging tracer. Our findings indicate that imaging-tracer distribution was strongly influenced by the anatomical and physiological properties of the target tissue and that such variability could be effectively predicted from DTI. Using our computational method, we show that parameters that govern fluid flow could be accurately assessed by DTI-MR imaging and used to provide highly accurate patient-specific predictions of drug distribution that may be useful for catheter placement and infusion planning.
Patients 18 years of age or older with a recurrent or progressive and resectable supratentorial MG (WHO grade III or IV) and a KPS score 70 were eligible for this study. A solid contrast-enhancing tumor nodule 1.0 cm and 5.0 cm was required. Patients enrolled also had to have completed external beam radiation therapy 8 weeks prior to study entry and recovered from toxicities of prior local therapies. Patients were excluded if they had signs of impending cerebral herniation, multifocal disease, tumor crossing the midline, or subependymal or leptomeningeal spread. The Duke Institutional Review Board (4774-03-4R0) and the Food and Drug Administration (BB-IND-8959) approved the protocol. Informed consent was obtained after the nature of all procedures was explained.
The study had a two-stage design. Patients enrolled in stage 1 received a combined preresection and postre-section continuous infusion of 123I-HSA coinfused with cintredekin besudotox. Patients in stage 2 received the postresection continuous infusion only. Patients enrolled in stage 1 underwent a stereotactic biopsy to confirm the existence of viable MG before stereotactic placement of two infusion catheters. At least one catheter was always placed into the contrast-enhancing component of the tumor. Before resection, cintredekin besudotox was infused at a concentration of 0.5 μg/ml for 96 h in a fixed total volume of 51.8 ml at a total infusion rate of 0.540 ml/h divided by the number of catheters placed. A craniotomy for tumor resection was performed 15 ± 7 days after the end of the preresection infusion with stereotactically guided, postresection, intraoperative placement of 1–3 infusion catheters into parenchyma surrounding the resection cavity.
In the postresection setting, cintredekin besudotox was infused at a concentration of 0.5 μg/ml over 96 h in a fixed total volume of 72.0 ml at a fixed total infusion rate of 0.750 ml/h divided by the number of catheters placed. Patients enrolled in stage 2 did not have a pre-resection infusion but rather a craniotomy followed by a postresection infusion identical to the postresection infusion used in stage 1 except that postoperative catheter placement occurred 3–7 days after resection and was performed through a small burr hole. Open-ended, barium-impregnated silicon infusion catheters with a 1-mm inner diameter and a 2-mm outer diameter were used (Vygon Neuro, Valley Forge, PA, USA). Guidelines outlined in the protocol for catheter placement were as follows:
Other suggested target selection criteria were as follows:
Each catheter positioning was scored by an observer blinded to the single-photon emission computed tomography (SPECT) and simulation results on a three-point scale using these criteria based on imaging within 24 h of placement. In addition, each catheter was also scored according to the criteria outlined for the phase III PRECISE (Phase 3 Randomized Evaluation of Convection-Enhanced Delivery of IL13-PE38QQR Compared to Gliadel Wafer with Survival Endpoint in Glioplastoma Multiforme at First Recurrence) trial of cintredekin besudotox (Table 1).
Cintredekin besudotox is a recombinant chimeric protein consisting of a genetically engineered, mutated, and truncated form of the cytotoxic Pseudomonas aerugi-nosa exotoxin fused to interleukin 13. The full sequence encoding cintredekin besudotox has been described.38 Prior to delivery, cintredekin besudotox was diluted with 0.2% HSA (Plasbumin-25; Bayer, Elkart, IN, USA) in 0.9% saline. For the first 48 h of each infusion, 123I-HSA was used as a surrogate for imaging the cytotoxin because the maximally tolerated dose is too small to radiolabel and image the cytotoxin directly. The HSA was purified to homogeneity by ion-exchange high- pressure liquid chromatography and radiolabeled with 123I (MDS Nordion International, Vancouver, Canada) by using a modified iodogen method with a target specific activity of 80 mCi/10 mg. 123I-HSA was chosen because its size, shape, and molecular weight are similar to those of the cytotoxin, HSA forms an otherwise essential component of cytotoxin drug formulations, and the concentrations of cytotoxin currently used in clinical studies are insufficient for direct labeling with sufficient radio-nuclide for imaging. In addition, recent work by Murad et al.39 has shown in well-controlled animal studies that labeled albumin does precisely track the distribution of cintredekin besudotox.
Brain MRI with unenhanced and contrast-enhanced T1 weighted (repetition time [TR] = 22, and echo time [TE] = 7), T2 weighted (TR = 6200, TE = 123), and DTI (six-direction 3-mm-thick contiguous slices, b value = 1000, TR = 8800, TE = 80) was obtained before each catheter placement to provide input data for the simulation algorithm. MR scans were obtained on a 3T scanner (Siemens Medical Systems, Erlangen, Germany). SPECT scans with a three-head scanner (Trionix Research Laboratory, Twinsburg, OH, USA) fitted with two Triad LESR (low-energy super high resolution) fanbeam collimators and a precise pinhole collimator were then obtained 24 and 48 h after infusion initiation to evaluate 123I-HSA distribution. The volume of distribution (Vd) was subsequently determined by a threshold pixel method that has proven accurate at our institution for calculating the volume of small spheres ranging in size from 1.3 to 5.3 cm3 in a brain phantom model.40,41 The Vd was based on the volume depicted by the SPECT at 50% of the maximal signal value.
The software algorithm was applied retrospectively in all cases by an observer blinded to the results of the SPECT imaging. Before simulating the actual fluid distribution, the surgical planning software first delineates fluid-filled volumes and surfaces, such as sulci, resection cavities, and ependymal surfaces by using a T2-weighted MRI data set because the resolution of clinically obtainable DTI data sets is currently too low to define these small anatomical structures. This is done using a three-dimensional (3D) ridge-filtering method. The ridge-filtering method is based on a local second-derivative operator that is maximized at thin peaks in the T2-weighted input image. This filter is effective at locating most sulci (if they are visible in the underlying image). However, other sharp boundaries, as may be found in areas of significant edema, can confound the algorithm. To prevent misclassifications that may result in these areas, the work flow was modified to include a preemptive step consisting of the manual segmentation of the edematous brain areas. To detect cavities and sulci, the pore fraction computed from the MR-DTI scan is also used. It is assumed that cavities exist where the pore fraction is estimated to be close to one. These methods are jointly referred to as sulcus detection.
When running the sulcus-detection algorithm, the software first uses the infusion flow rate and catheter dimensions to estimate the length of fluid backflow along the catheter track. Within this estimated length, the software then checks each catheter trajectory for the presence of a segmented surface or cavity. If a surface is detected, the software brings up a dialogue box containing a warning regarding a potentially poor catheter trajectory that is at risk for failing to produce intraparenchymal distribution of the infusate. This enables the user to return to the planning mode and check the catheter trajectory for potential repositioning (Fig. 1A). Only after the user accepts the trajectories at this stage is the fluid distribution actually simulated as described below.
Morrison et al.13 describe the rate of change of drug concentration per unit tissue volume c as an approximation from a sum of diffusion, changes caused by convection, and losses:
In this equation, D stands for the diffusion tensor of the drug molecule in the interstitial space, v describes the velocity of the interstitial fluid, is the pore or interstitial volume fraction, and k accounts for the irreversible metabolism losses and for the disappearance through capillaries. We have been engaged in a study aimed at solving such equations in a subject-specific manner, that is, where the aforementioned distributed parameters, or fields, are obtained for a specific individual (Chen et al., submitted; Raghavan et al., submitted, U.S. Patent 6,549,803, and U.S. Patent 6,464,662). The equation is solved with appropriate boundary conditions for an individual brain by obtaining and estimating the parameters (D, v, , and k) from MRI and from the literature.
The velocity in the interstitial space is computed by applying D’Arcy’s law, which relates the interstitial pressure gradient and the interstitial fluid velocity linearly, the coefficient being the hydraulic conductivity tensor,
Finally, by combining Equation 2 with an expression for the differential conservation of water, the following equation is obtained:
where Lp is the capillary hydraulic conductivity governing the rate of net flow of water across capillary membranes and s is the capillary area per unit tissue.
The main parameters, D, K, and are computed from MRI. The water self-diffusion tensor field, Dw, is obtained from MR-DTI. From this, the diffusion of the drug molecule is estimated from a simple scaling law and the weight of the molecule. Dw is also used to estimate the porosity, , and finally a map of the hydraulic conductivity tensor, K, is obtained from Dw and via cross-property relations. We thus obtain patient-specific 3D maps of these quantities, which are used as input for the computer simulation algorithm as described by Chen et al.42
The simulation begins by solving Equation 3 for the pressure field related to the infusion. The required boundary condition in this partial differential equation for the pressure is obtained by computing the pressure profile along the catheter shaft based on a poroelastic model of backflow, first described by Morrison et al.13 Given the pressure along the catheter shaft, Equation 3 is solved, and then the fluid velocity field v is obtained by using Equation 2. Finally, by using this estimate for v, Equation 1 is solved. The result is thus a patient-specific map of fluid concentration at any desired time point during or after the infusion. In surgical planning software (Therataxis, Baltimore, MD, USA; and BrainLAB, Feld-kirchen, Germany), this result can then be displayed as a 3D overlay on the anatomical MRI scans, enabling the physician to assess whether the volume covered with the infusion given a set of catheter trajectories will be satisfactory. The software assists in the optimization of the planned trajectories by enabling the simulation to be run at different catheter locations. Computational time for the fluid simulation depends on the simulation resolution and, with the software implementation evaluated in this study, is in the range of 3–10 min.
The simulation algorithm is not currently designed to handle the effects of large local variations in blood- brain-barrier permeability that may be seen within unresected tumor tissue, although we believe that, by incorporation of dynamic imaging of contrast enhancement, this may be possible. The simulation algorithm was, therefore, evaluated only on catheters placed in the postresection setting.
For all catheter trajectories evaluated, the sulcus-detection algorithm was run first. For all trajectories that were not identified as problematic by this algorithm, the fluid distribution simulation was performed.
The accuracy of the simulation in predicting the Vd was evaluated by dividing the concordant volume (volume for which the simulation [SIM] and the SPECT were in concordance) by the sum of all volumes:
where (SPECT > SIM) stands for the volume where the SPECT signal outline was larger than the simulation signal outline and (SPECT < SIM) describes the SPECT signal that was not covered by the simulation (Fig. 1B).
The accuracy of the simulation in predicting the geometric distribution of the infusate was evaluated by measuring the maximum distance between the windowed SPECT border and the simulation border at the 50% isodose level. For this measurement, the slice with the largest distance between the simulation and the SPECT signal was always used (Fig. 1C).
The purpose of the simulation software is to support clinicians in identifying catheter trajectories unlikely to provide drug delivery to the desired anatomical distribution, and for suitable trajectories, to estimate the expected volume and geometric distribution of the tissue covered by the infusate. Thus, for each infusion catheter evaluated, the software was graded as “clinically useful” if it identified catheter trajectories that failed to deliver any drug into the desired anatomical region or if it provided a fluid flow simulation with volume match of 50% or greater or an in-plane deviation of 10 mm or less.
Fisher’s exact test was used to assess the association between the actual presence along the catheter trajectory of deep sulci or other anatomical variations that caused the infusate to leak into the subarachnoid cere-brospinal fluid (CSF) space and the algorithm’s prediction. The sensitivity, specificity, positive predictive value, and negative predictive value for the algorithm’s prediction were calculated, including a 95% CI. A one-sample Student’s t-test was used to test the one-sided null hypothesis that the mean Vd concordance is less than 50% (H0: Vd < 50%; and H1: Vd 50%). A one-sample Student’s t-test was also used to test the null hypothesis that the mean maximum in-plane deviation is less than 10 mm as demonstrated by SPECT at the 50% isodose level for both (H0: deviation < 10 mm; and H1: deviation 10 mm).
Eight patients were enrolled at Duke University Medical Center between July 22, 2003, and January 28, 2004. Seven patients received infusions from a total of 21 catheters, whereas one patient, enrolled in stage 2, had tumor resection but experienced postoperative edema precluding catheter placement and postresection infusion. The median age was 46.5 years (range, 19–62). The mean KPS score was 90 (range, 80–100). Four patients (50%) had an initial histopathologic diagnosis of GBM, whereas three (37.5%) had anaplastic astro-cytoma, and one (12.5%) had a low-grade astrocytoma that underwent anaplastic transformation to a GBM. All patients had previously been treated with surgical resection, external beam radiation therapy, and at least one cytotoxic chemotherapy.
All four patients in stage 1 of the study received the study drug. Two patients received the planned preresec-tion and postresection cintredekin besudotox infusions, one patient only had the preresection infusion because of rapid tumor progression that precluded resection and postresection infusion, and one patient had interruption of the preresection infusion approximately 24 h after initiation because of SPECT imaging evidence of drug distribution only into the CSF compartment. This patient subsequently had resection and the postresec-tion infusion without incident. In stage 2, three patients received the planned postresection infusion, while one patient after tumor resection, but before drug infusion, experienced postoperative edema precluding catheter placement and postresection infusion. Other events related to catheter placement were only grade 1 or 2 and included headache and postprocedural pain (29%), increased intracranial pressure (29%), wound-related complications (29%), transient aphasia (14%), and transient hemiparesis (14%). The only drug-related adverse event was nausea (grade 1) in two patients.
A total of 21 catheters were placed in this study. Since the software under review is intended for use in the postresection setting only, and because CED is currently used most frequently to provide infusions after tumor resection, 7 of the 21 catheters placed intratumorally were excluded from our evaluation. Of the 14 catheters evaluated, seven (50%) failed to produce drug distribution in the desired anatomical region, as evidenced by SPECT. Of these seven catheter trajectories, six resulted in leak of the infusate into the subarachnoid CSF space. This occurred either because they had their tips located in or near a resection cavity or ependymal surface or because the catheter trajectory crossed a pial surface within a deep sulcus (Fig. 2). The remaining catheter of the seven produced an unexpected and unpredicted distribution within the parenchyma, but along a path at nearly right angles to the catheter trajectory. In this case, we believe the infusate distributed along a catheter tract left over from a previous infusion that was intersected by the trajectory of the studied catheter (Fig. 3). Of the seven catheters that produced such ineffective infusions, the sulcus-detection algorithm correctly identified five as problematic, giving an overall sensitivity of 71.4% (95% CI, 29%–96%) (Table 2) (p = 0.021).
Without manual segmentation of edematous areas, the software algorithm misclassified borders of edema as sulci in four of the 14 catheters evaluated (false positive). However, after incorporation of manual segmentation of edematous regions into the software algorithm and segmentation of these areas by a user blinded to the software simulation results, for the seven catheters where no sulci or cavities were present along the trajectory, the software algorithm never identified other structures as sulci. The specificity of the sulcus-detection component overall then was 100% (95% CI, 59%–100%) (Table 2) (p = 0.021). This significant improvement in specificity with the addition of the manual edema segmentation task indicates that this manual approach is a good solution to this problem. However, an automated approach to this problem is being developed.
Of the 14 catheters evaluated, five trajectories were flagged by the sulcus-detection algorithm as invalid as previously described and thus were not further simulated (see Fig. 1A). Of the remaining nine catheters, infusion from one was delayed such that the signal from the 123I-HSA had decayed to a point below detection. Therefore, eight infusions visualized by SPECT were available for comparison with the simulation results for concordance of the Vd (Fig. 1B).
For six of eight usable SPECT volumes, the concordance between the Vd of 123I-HSA as measured by SPECT at the 50% isodose and the Vd predicted by the simulation was more than 50% (Table 3). The mean Vd match was 65.75% (95% CI, 52.0%–79.5%) (p = 0.028). Of interest is the finding that a 100% match is shown for catheter 2 in patient 108. This match resulted from the sulcus-detection algorithm not detecting the cavity into which this catheter was placed; the fluid simulation, however, still returned an empty volume as a result, thus correctly accounting for the fluid loss into such cavity. Conversely, catheter 1 in patient 102 traversed a prior catheter tract, as previously described, leading to preferential egress of the infusate down that catheter tract and a poor volume match (Fig. 3). The simulation algorithm failed to recognize this thin catheter tract as a potential avenue of misdirected infusate egress on the DTI and therefore failed to provide an accurate infusion Vd.
For all eight infusion volumes available for evaluation, the maximal distance between the edge of the simulated fluid distribution and the edge of the SPECT infusion at the 50% isodose level was measured (Fig. 1C). Table 4 shows the results for all of these in-plane deviation distance measurements. For six of eight infusion volumes, the in-plane deviation is less than 10 mm. The mean maximum in-plane deviation between the simulation and the SPECT-imaged distribution is 8.5 mm (95% CI, 4.0–13.0 mm) (p = 0.266). As expected, catheter 1 in patient 102 provides the worst result, as previously described.
To evaluate the potential clinical utility of the algorithm, 13 separate infusions were reviewed (Table 5). The clinical utility outcome criterion combines the results of both the sulcus-detection algorithm and the fluid flow simulation algorithm, and therefore can be regarded as the overall usefulness of the software version available at the time of this evaluation (fall 2005). According to the criteria outlined a priori and described in the Patients and Methods section, the results suggest that the software algorithm provides clinically useful information for 11 (84.6%) of the 13 infusions. In five infusions, clinical utility was based on the sulcus-detection algorithm, and in six cases, it was based on volume matches and in-plane deviation measurements. For the two infusions where the software appeared to have no clinical utility, the algorithm failed in both the Vd and in-plane deviation evaluations. One of these was the catheter with a trajectory that crossed a preexisting catheter track.
To evaluate whether adherence to the catheter positioning guidelines employed in this trial or to those developed for the phase III PRECISE trial of the same agent is sufficient to result in adequate infusate distributions, we evaluated the catheter positioning in this trial against those criteria. Not all catheter placements in this study fulfilled the guidelines outlined in the protocol for this study as previously described, nor did they all adhere to the guidelines used in the phase III PRECISE trial (Table 1). However, failure to meet these guidelines was not always the reason for poor distribution (Table 6). In contrast, of the nine catheters that failed to meet more than one of the guidelines outlined for this trial, five (55.6%) still produced significant infusate distribution. Similarly, of the four that met only one of the criteria outlined for this protocol, three (75%) produced significant intrapa-renchymal distribution. No catheter failed to meet all criteria. Conversely, of the 10 catheters that met at least two of the guidelines outlined for this trial, six (60%) failed to produce any significant intraparenchymal drug distribution.
A further evaluation of catheter positioning in this trial according to the less restrictive guidelines established for the phase III PRECISE trial of cintredekin besudotox also fails to explain all poor distributions. According to the PRECISE guidelines,43,44 10 (71%) of 14 catheter placements in this trial obeyed all guidelines, were classified as “optimally placed” and received the highest score of two points. Of these 10 optimally placed catheters, however, only six (60%) produced a significant intraparenchymal distribution as evidenced by SPECT. These guidelines do appear to have some predictive value, however, as three (75%) of four catheters with less than optimal scores produced no intraparen-chymal distribution.
The findings presented here provide evidence that MR-DTI images contain valuable patient-specific information that could be effectively exploited to assist in the optimal placement of intracerebral catheters for CED. In this retrospective study, we found that a pilot software algorithm that incorporates patient-specific data derived from MR-DTI could provide clinically useful information regarding the location, volume, and geometry of distribution of a radiolabeled imaging tracer delivered by the novel technique of CED in patients with MG. These findings are important because surprisingly few studies using CED monitor drug infusion, and failure to deliver drug to all areas of tumor infiltration will clearly constrain the potential of any agent being delivered.
One remarkable observation in this study was that 50% of the catheters placed failed to produce significant intraparenchymal drug distribution despite the majority being classified as optimally positioned according to the best available criteria. These findings suggest that other factors such as tissue heterogeneity and brain pathology, in addition to those considered by the guidelines, may play a role in drug distribution. The software algorithm described here for simulating these infusions takes into consideration some of these factors, and this may account for some of its usefulness.
Although the current algorithm provides useful information that may assist with the placement of catheters for drug delivery by CED, this study also identified a number of shortcomings in this approach that need further development. First, although the specificity of the sulcus-detection algorithm was found to be very high (100%), in part because of the manual edema segmentation that avoids misclassification of edematous areas as fluid-filled cavities, the sensitivity of this algorithm needs to be improved. In general, specificity will be less important than sensitivity because end users will have a wealth of anatomical information that they can use to exclude false positives. Second, although the overall match for Vd at the 50% isodose level was satisfying, volume mismatches and planar deviations were identified. Although these deviations may be due to an inaccurate simulation, they may also be attributable to the relatively low resolution of SPECT imaging or to the comparably high spatial distortion of currently available MR-DTI input data. Third, differences in the distribution of the inert tracer 123I-HSA used in this study and the actual therapeutic drug being delivered will need to be formally assessed, as will potential differences with therapeutic molecules of different molecular weights and binding kinetics. Still, where large infusate leaks are predicted by the software, it is reasonable to assume that these leaks would be the predominant force influencing drug delivery, and thus the simulation still provides useful information in these cases. Finally, in future studies, the parameters governing infusion into or near solid tumor masses should also be studied.
One foreseeable application of this software is its use in elucidating some potential general principles of CED by testing different infusion scenarios. For example, from a mock simulation (Fig. 4), one can appreciate the difficulty in attempting to provide drug coverage to a 2-cm margin around a resection cavity—the area at highest risk for tumor recurrence—with a limited number of catheters. Such limitations do not appear to have been appreciated in animal studies that employed smaller brains or in existing clinical trials.
In conclusion, this software algorithm shows signifi-cant promise for enhancing CED of therapeutic macromolecules to the human brain for neoplastic and other conditions. As imaging techniques evolve, the input data for the algorithm should gain in spatial accuracy and resolution, and it is estimated that this will automatically produce better simulations. Other potential routes for improving input-data quality may be the combination of information from various imaging modalities, for example, dynamic imaging of contrast enhancement, to enhance the information used in the algorithm and enable better estimates of drug efflux from the brain.
This research was supported by NIH/NCRR K23 RR16065 (J.H.S.), NIH/NCI R01 CA097611 (J.H.S.), 2P50-NS20023 (D.D.B. and J.H.S.), 5P50-CA108786 (D.D.B. and J.H.S.), BrainLAB, and NeoPharm. Experimental data were acquired by using shared instrumentation funded by the National Center for Research Resources of the National Institutes of Health (S10 RR15697). These studies were conducted as part of collaboration between the Food and Drug Administration and NeoPharm under a Cooperative Research and Development Agreement (CRADA). The views in this article do not necessarily reflect those of the Food and Drug Administration. This research was supported in part by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.
The authors also acknowledge the contributions of Dr. B. H. Joshi, Lisa Tansey, Kim Greer, Neil Petry, Sharon McGehee, Denise Lally-Goss, Amy Grahn, Dr. Jeffrey Sherman, Dr. Andreas Hartlep, and Michelle Smith.