Automated Volumetric Growth Plate Measurement Using Magnetic Resonance Imaging for Monitoring Skeletal Toxicity in Children Treated on Investigational Drug Trials

The focus of anticancer drug discovery and development has shifted from cytotoxic to newer classes of drugs that selectively target protein and signal transduction pathways that are directly involved in the pathogenesis of cancer (1). Many of the pathways important for tumor pathogenesis may also play a vital role in normal growth and development. Thus, children may have potential toxic side effects from these agents that are unique from those in adults. The skeletal system is one of the target organs that may be affected by a variety of targeted agents such as angiogenesis inhibitors (2), hedgehog inhibitors (3), retinoids (4, 5), and multi-tyrosine kinase inhibitors (6–9).

Potential toxic effects on the skeletal system have been most frequently described with the class of angiogenesis inhibitors. Agents that inhibit the VEGF ligand or VEGF receptor (VEGFR), such as bevacizumab and sorafenib, have shown antitumor effects in adults and are currently FDA-approved for treatment in some adult cancers. However, the use of these agents may also result in significant toxicity to a number of target organs that rely on angiogenesis for maintenance and development, such as longitudinal bone growth (2).

Longitudinal bone growth occurs at the growth plate, a highly specialized cartilaginous structure situated near the ends of tubular bones and vertebral bodies (10). VEGF is an essential coordinator of chondrocyte death, chondroclast function, extracellular matrix remodeling, angiogenesis, and bone formation at the growth plate (11). Inhibition of VEGF signaling affects bone growth by thickening the growth plate, which results from expansion of the hypertrophic zone. These effects have been attributed to delayed vascular invasion of the growth plate resulting in a reduced rate of hypertrophic chondrocyte apoptosis (12). In preclinical models, the changes in growth plate thickness can be morphometrically quantified, revealing a dose-dependent increase in the growth plate area of up to 481% (13). Inhibition of VEGF also impairs trabecular bone formation, as anti-VEGF treatment showed reduced length and number of primary trabeculae in the long bones of juvenile mice but not in mature animals (11). These changes were reversed within 2 weeks of cessation of the anti-VEGF treatment. Normal angiogenesis was restored and rapid reversal of all growth plate changes occurred (11). Reversible bone growth abnormalities have also been observed after the administration of sorafenib in preclinical models (14).

In the clinical setting, one case report describes the development of punched-out metaphyseal bone lesions in an infant with cutaneovisceral angiomatosis treated with bevacizumab (15). The lesions were noticed after 4 doses of bevacizumab administered at 2-week intervals and they resolved after 2 months of stopping the agent. In the phase I study of vandetanib during and after radiotherapy in children with diffuse intrinsic pontine glioma, one of 30 patients who underwent follow-up studies during therapy had a small area of premature fusion of the cartilaginous growth plate (∼2% of total area), shown by an MRI of the knee (16). The authors concluded that changes in growth plate were uncommon after the use of vandetanib, but lack of long-term toxicity could not be determined due to the short follow-up in most patients. There are no other known reports in the literature of growth plate changes or growth retardation in children treated with angiogenesis inhibitors. Traditional pediatric phase I cancer trials, where children are on treatment study for a median of approximately 1 to 2 months (17), may not allow for sufficient follow-up time, and/or methods used may not be sensitive enough to detect any small changes. Skeletal toxicity has not been described in adults receiving treatment with angiogenesis inhibitors and may exclusively develop in children whose growth plates are undergoing active remodeling.

Other targeted anticancer agents have reported side effects on the skeletal system in preclinical models. Imatinib mesylate is a multi-tyrosine kinase inhibitor, which is FDA-approved for treatment of children with Philadelphia chromosome-positive chronic myelogenous leukemia that is resistant to IFN-α therapy or recurrent after bone marrow transplantation. Treatment with imatinib in rats resulted in decreased growth plate thickness and inhibition of longitudinal bone growth (7). Preclinical studies in juvenile mice of smoothened (SMO) inhibitors, downstream targets of the hedgehog pathways, showed rapid differentiation of chondrocytes leading to dramatic expansion of the hypertrophic zone (3). Removal of the drug quickly restored the pathway activity, but precipitated bone mineralization and premature fusion of the growth plate. SMO inhibitors are currently being developed as potential therapies for several cancers including pediatric medulloblastomas.

The evaluation for potential skeletal toxicity in children enrolled on clinical trials with these classes of molecularly targeted agents is thus critical. The primary objectives of phase I trials are to describe the toxicity profile and to recommend a phase II dose for new agents under evaluation. Most toxicities in phase I cancer trials are graded according the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events (CTCAE). The CTCAE version 3.0 includes a grading mechanism for bone-age alterations, which only includes grade 2 toxicity for ±2 SDs from the normal bone age, and grading is also available for reduction in growth velocity. Both criteria are not included in the recent CTCAE version 4.0, and no other specific grading system to monitor for skeletal toxicity and growth plate thickness is described. The Children's Oncology Group has incorporated serial scanograms or anteroposterior radiographs of the tibial growth plate in their phase I trials with angiogenesis inhibitors or other agents with potential for skeletal toxicity. These methods may allow for gross visual changes over time, but as the growth plate is only millimeters in thickness, subtle changes may be difficult to quantify to assess for toxicity, and no grading system is provided.

We developed and validated a method that uses noncontrast MRI to semiautomatically measure the growth plate volume with the goal of allowing for reproducible and sensitive detection of the growth plate toxicity. This method can be used to detect small-volume changes in serial evaluation of growth plates in children who enroll into clinical trials with investigational agents with potential to affect growth and development.

A knee MRI obtained from 20 subjects who were enrolled on a screening evaluation protocol for potential enrollment in a pediatric clinical trial open at the NCI, Pediatric Oncology Branch from February 2007 to November 2009 were used for validation. All subjects had MRIs of their right knee done except for 2 subjects who had left knee MRIs because of tumors obstructing the site of the growth plate at the right knee. To document intrasubject reproducibility, 5 subjects enrolled on the NCI Neurofibromatosis Type 1 (NF1) Natural History Study from December 2010 to February 2011 had 2 sequential unilateral knee MRIs done on the same day in 2 separate MRI sessions. These scans were only done in subjects who did not require sedation and were scheduled to undergo MRI scanning for evaluation of their disease. This method was used in the phase I trial of the angiogenesis inhibitor cediranib (18) to longitudinally monitor growth plate changes in children and adolescents with refractory solid tumors. All study protocols were approved by the NCI institutional review board. All subjects had consented to the MRI and volumetric MRI analysis of their knee.

A software program to semiautomatically measure the growth plate volume of the knee using MRI was developed by Jeffrey Solomon (Medical Numerics, Inc.). This method is based on a similar program he developed for the semiautomated volumetric analysis of plexiform neurofibromas in subjects with NF1 (19). MRI data of the knee were acquired with a standardized T₁-weighted sagittal sequence, without the use of intravenous contrast, from a 1.5T General Electric Signa MRI scanner based on the following MRI specifications: echo-train length, 5; repetition time, 500 to 600; echo time, min full; slice thickness, 3 mm; bandwidth, 20 kHz; interslice gap, 1.5 mm; field of view, 18; frequency, 512; phase, 256; number of excitations, 1; phase field of view, full; and no fat saturation. Growth plate volume was determined using the MEDx software platform (Medical Numerics, Inc.). The method is based on pixel intensities within the region of interest and conducts a Bayesian probabilistic analysis, using a Gaussian mixture model, to classify the pixels into 3 different classes. The steps for the volumetric analysis are outlined in Figure 1. The growth plate appears dark compared with the adjacent bone structures (Fig. 1A). The region of interest including the low signal intensity growth plate and some higher signal intensity adjacent bone is manually outlined on each MRI slice (Fig. 1B). The software program then automatically displays the class of interest, the growth plate, in red (Fig. 1C) on each slice and calculates the volume of all the red pixels.

The MRI slice with the most complete view of the posterior cruciate ligament was selected to measure the growth plate thickness. A horizontal line spanning the femoral growth plate was drawn. Another line measuring half of the preceding horizontal line was drawn to determine the midway point of the growth plate. The perpendicular one-dimensional (1D) femoral growth plate thickness was measured at the midway point. An estimate of growth plate diameter was measured on the same MRI slice using the measurement from the anterior edge of the metaphysis of the femoral bone to its posterior edge (Fig. 2).

To determine the intra- and interobserver variability and reproducibility of the volumetric growth plate and 1D thickness measurements, 2 physicians applied these methods to the 20 subjects with knee MRIs. Each physician calculated a femoral growth plate volume and 1D femoral growth plate thickness from each knee MRI on 3 different days. The tibial growth plate was also calculated in a subset of MRIs (n = 5). The mean ± SD and coefficient of variation of the 3 femoral growth plate volumes and 1D measurements were calculated for each knee MRI for each reader to evaluate intraobserver reproducibility. For interobserver variability, the mean difference was calculated by taking the mean value of observer 1 minus the mean value of observer 2 divided by the total mean value. A linear regression was conducted for correlation between the 2 observers. A Bland–Altman plot was drawn to analyze the concordance between the 2 observers. Pitman's test of difference in variance was used to test the concordance. In addition, each subject's mean growth plate volume, thickness, and diameter measurements were correlated to age, body surface area (BSA), height, weight, time of scan (Spearman's ρ), and Tanner stage and gender (Wilcoxon rank sum). Bone age was not analyzed in the majority of children, and this information could, therefore, not be included in our analysis. To evaluate intrasubject reproducibility of the volumetric measurements, the volumes of the 5 subjects who had 2 sequential scans done on the same day were compared using the Wilcoxon signed-rank test. The longitudinal data of the volumetric data from the cediranib study are presented in a descriptive manner. Statistical analysis was carried out using Stata version 9.0 (College Station, Texas).

The characteristics of the 20 subjects who had knee MRIs done for volumetric analysis of their growth plate are described in Table 1 along with their measured femoral growth plate volume, 1D thickness, and diameter on knee MRI. There was a good diversity of ages (median: 12 years; range: 5–17 years) and there were 10 females and 10 males. Thirteen subjects were classified as Tanner stage I. The median height was 138 cm (range: 105–172 cm) and weight was 28 kg (range: 17–70 kg).

The mean volumetric and 1D thickness measurements of each femoral growth plate from the 20 knee MRIs by each observer are detailed in Table 2. The mean growth plate volume was 5,122 mm³ (SD = 2,009). For the volumetric analysis, the intraobserver variability was very low, as shown by the coefficient of variation for the 2 observers, which ranged from 0.2% to 6.1%. The correlation coefficient between observer 1 and observer 2 was 0.99 (Fig. 3A). The mean percentage difference between the 2 observers was 3.6%. The concordance between observers was good as evidenced by the Bland–Altman plot (Fig. 3B), where the measurements were well dispersed around the mean difference, which was −1.8 mm³, and all but one measurement were within the limits of agreement. Pitman's test of difference in variance showed no independence (r = −0.061; P = 0.8). In contrast, the intraobserver variability was greater for the 1D growth plate thickness as shown by the coefficient of variation for the 2 observers, which ranged from 2.3% to 53.1%. The interobserver agreement showed poorer correlation (r² = 0.27; Fig. 3C). Although the mean difference was near zero and most measurements were within the limits of agreement in the Bland–Altman plot, the values were well dispersed between observers leading to a wide confidence interval (Fig. 3D).

Tibial growth plate volumes were analyzed in a subset of subjects (n = 5) to evaluate whether similar or improved results compared with those of the femoral growth plate could be observed. The results were similar to those of the femoral growth plate with low intraobserver variability, with the coefficient of variation for one observer ranging from 0.5% to 10%.

Five subjects enrolled in the NF1 natural history study had 2 sequential knee MRIs done on the same day in 2 separate MRI sessions. The median age was 14 years (range: 12–15 years), with 2 females and 3 males. There were no statistically significant differences seen between the volumetric measurement of the first scan and the second scan (P = 0.5; Fig. 4).

Five subjects out of the 18 enrolled in the phase I cediranib study had more than 2 knee MRIs done, one at baseline and at least 1 other at a subsequent cycle (one cycle = 28 days). The median age for 5 subjects was 14 years (range: 11–15 years) with 4 females and 1 male. The dose of cediranib ranged from 12 to 17 mg/m²/d and 4 subjects received 4 cycles or less of therapy and 1 subject received 27 cycles of therapy. For 3 subjects who had baseline and post-cycle 2 knee MRIs evaluated, significant changes were neither observed in their femoral growth plate volume or growth plate diameter nor did they grow in height during this time. One subject (14 years old) had knee MRIs at baseline, post cycles 2 and 4 of therapy with femoral growth plate volumes of 3,701, 3,544, and 3,751 mm³, respectively, and her height increased by 0.8 cm during this time. The growth plate diameter also remained unchanged measuring at 29.2, 29.9, and 29.8 mm. The subject (11 years old) who received approximately 2 years of therapy had 7 knee MRIs done during this period. Her initial growth plate volume was measured at 4,360, and the final measurement was 4,133 mm³ without significant changes in between with absolute differences of volume from baseline of −96 to 643 mm³ (range: 2%–15%). Of note, her post-cycle 8 therapy scan was not included because of poor image quality. Her initial growth plate diameter was 28.3 mm and final was 28.1 mm, with no meaningful differences seen in measurements of all 7 knee MRIs. She grew 1.8 cm during this period, and her course was complicated by anorexia and weight loss attributable to both therapy and disease.

Baseline growth plate volume was correlated to baseline characteristics of the subjects (n = 20). There was a wide, scattered distribution of growth plate volume resulting in only a moderate correlation to a subject's anthropometric measurements (Spearman's ρ = 0.4, 0.5, and 0.5 for height, weight, and BSA, respectively). When the analysis is limited to prepubertal Tanner stage I children (n = 13), there appears to be a stronger positive correlation between body size and growth plate volume. The correlation coefficient (ρ) for height, weight, and BSA was 0.5, 0.8, and 0.7, respectively. This correlation was statistically significant for weight and BSA (P < 0.01). Prepubertal children (Tanner stage I) had, on average, larger growth plate volumes than children staged greater than Tanner stage I (6,043 vs. 4,626 mm³; P = 0.2), but this was not statistically significant. The males also had, on average, larger growth plate volumes than females (5,921 vs. 4,323 mm³; P = 0.05). The males were, on average, taller and heavier than the females in this study, which confounds this finding. No statistically significant associations to age or timing of the MRI scan and growth plate volume were observed. 1D growth plate thickness and subjects' anthropometric measurements appear to be loosely negatively correlated, although none were statistically significant. The growth plate diameter does appear to be positively correlated to subjects' age, weight, height, and BSA with correlation coefficient (ρ) 0.8, 0.8, 0.9, and 0.8 respectively (P < 0.01).

New toxicities have emerged with the advent of biologically targeted anticancer agents compared with cytotoxic chemotherapy. Clinical researchers in early drug development must be able to grade and manage novel toxicities to determine tolerability and dose of new agents. Thus, methods to define and monitor new drug toxicities are important goals for all cancer therapies.

There are many novel investigational agents currently in pediatric clinical trials for refractory cancers or tumor-predisposition syndromes such as NF1. The effect of these agents, mainly angiogenesis inhibitors, in preclinical models resulting in growth plate thickening due to retention of the hypertrophic chondrocytes (2, 11, 13) has raised the question of how this potentially unique toxicity can be monitored meaningfully on clinical trials in children. This is of particular importance to children who will be receiving chronic treatment with these agents. For example, children with NF1 and plexiform neurofibromas remain on phase I trials for a much longer duration than children with refractory cancers (a median of 10 vs. 1 cycle, respectively; ref. 20). In addition, growth plate changes will be crucial to evaluate when these novel agents become incorporated into upfront cancer treatment trials in children.

No known gold standard exists for the measurements of the growth plate. The preclinical studies (11, 13) and quantitative analysis of human growth plates have been done with histomorphometric analysis (10), which would clearly not be possible in our study population. Previous studies with bevacizumab, a humanized VEGF antibody, used lower extremity scanograms and bone age (21) or knee X-ray films (22), and one study with vandetanib, a small-molecule inhibitor of multiple kinases including VEGFR2, used knee MRI (16) for growth plate evaluation. However, quantification of growth plate changes was not defined in these studies.

The MRI characteristics of the femoral growth plate, which is large and images differently from surrounding bone, allowed us to develop a semiautomated method of volumetric growth plate analysis, which can be used to accurately and consistently evaluate growth plate changes in children who enroll in clinical trials with investigational agents. The advantages of volumetric MRI analysis are that it does not require intravenous contrast administration, does not expose the subject to radiation, and requires only 15 minutes of scan time. As MRIs are increasingly being used for required disease evaluation, the knee MRI can be done at the same time as disease restaging, which is an important consideration in those children who require sedation for imaging. The automated volumetric growth plate analysis is easy to learn and do, and the analysis can be completed in ∼2 minutes. The tibial growth plate was also evaluated and showed consistent findings compared with the femoral growth plate such that this method would be adaptable to multiple bones.

Our study shows that the volumetric method is reproducible within observer and between observers with excellent correlation and concordance. As the growth plate is not uniform in thickness, MRI landmarks were used to consistently define the location for the 1D measurement. We were able to measure the 1D growth plate thickness in the same subjects with a total mean growth plate thickness of 2.1 mm. With such small measurements, changes of even 0.5 to 1 mm, likely due to technical limits, resulted in large variations within and between observers. For example, comparing interobserver measurement of subject 11, the difference between 3.2 and 1.1 mm is almost 3-fold. In comparison to 1D MRI measurements, volumetric measurements showed superior intra- and interobserver reproducibility and thus can be used to quantify small differences meaningfully.

This method showed good intrasubject reproducibility. When 2 sequential same-day scans were evaluated in 5 subjects allowing for volumetric measurement analysis without the confounding variables of time or potential drug effect, the volumetric measurements of the 2 scans showed no statistically significant differences. In addition, 3 subjects who were treated on cediranib therapy with femoral growth plate volumes measured at baseline and at 2 months, where there were no documented changes in height, showed no differences in between the 2 measurements.

Baseline growth plate volume may be correlated to anthropometric measurements such as weight and BSA, particularly in prepubertal children. This could be a potential bias when determining whether growth plate changes over time are due to changes in growth or drug effect. With increasing age and development, growth plate height decreases, as does the number of cells in the proliferative zone and the number of hypertrophic cells (10). This process continues throughout childhood and even during periods of rapid bone growth such as pubertal growth spurt (10). Thus, physiologically, growth plate thickness should decrease as height increases over time, and narrowing of the growth plate is most dramatic during mid- to late puberty. The diameter of the growth plate does appear to gradually increase over time as a child ages and grows. Thus, even in the presence of stable or decreasing growth plate thickness, physiologic appositional growth may contribute to an overall increase in growth plate volume. This may be why we see a stronger positive correlation between growth plate volume and body size in prepubertal children; however, we did not observe a statistically significant correlation of growth plate volume to Tanner stage or age.

This bias is minimized during a course of a clinical study as subsequent volumetric measurements would be made in comparison to each subject's baseline measurement allowing for our method to detect changes over time. In addition, one could measure the growth plate diameter as an estimate of appositional growth. The animal models showed a dramatic 4-fold increase from baseline of growth plate area after treatment with angiogenesis inhibitors (11). For the 2 subjects on the phase I cediranib study in children and adolescents who were followed for an extended period of time with increased documented height, minimal changes were appreciated in femoral growth plate volume from baseline. No changes were observed in the growth plate diameter during this time. From these observations, we concluded that these 2 subjects did not appear to experience growth plate toxicity due to cediranib. Although clearly preliminary due to a small sample size, another conclusion was that physiology alone does not appear to be a major contributor to growth plate volume changes during the interval of time for toxicity monitoring on drug trials. Thus, significant changes in growth plate volume without changes in appositional growth likely would be attributed to drug effect rather than physiologic change. Our method allows for quantification of growth plate volumes, thus allowing for sensitive detection of drug effects over time. It can be used to develop guidelines for specific grading of growth plate toxicity much like we do for toxicity monitoring of many laboratory measurements. 1D thickness measurements did not appear to be sensitive enough to quantify change because large variations in measurements were seen between observers probably due to the technical limits of measuring a growth plate that is only millimeters thick.

Although the age range of subjects in this study has good diversity, with the youngest subject being 5 years old, the subject population was limited by those who came to the NIH for potential treatment in investigational drug trials. It would be important to evaluate this in very young children, as this group may be at greatest risk for long-term sequelae. For future analysis, it would be useful and feasible to develop normative volumetric data from healthy children utilizing this method in a longitudinal prospective trial.

Volumetric analysis of the growth plate allows monitoring for only 1 aspect of potential bony toxicity, and additional modes of monitoring growth and bone metabolism should be considered in the design of studies with investigational agents in children. For example, in our ongoing phase I study of sorafenib for children with NF1 and plexiform neurofibromas, in addition to volumetric growth plate analysis, close monitoring of height with standard stadiometer measurements and other modes, such as tibial length and arm span, are included. To evaluate potential changes in bone turnover and metabolism, dual-energy X-ray absorptiometry scan, and laboratory evaluations for serum calcium, phosphorus, parathyroid hormone, osteocalcin, and bone-specific alkaline phosphatase have been incorporated to ensure comprehensive evaluation for potential bony toxicities.

In conclusion, MRI semiautomated volumetric growth plate measurement is a reliable and reproducible method that can be used to evaluate growth plate changes in children who enroll into clinical trials with novel, molecularly targeted agents with potential to affect skeletal development and growth. Growth and skeletal disturbances from these drugs are potential toxicities that may be unique to children whose growth plates are undergoing active remodeling. This method, together with other modes to evaluate bone and growth changes, may have utility for the monitoring of toxicity in clinical trials with novel agents in the pediatric population. Future evaluations in subsets of children receiving concomitant medications that may contribute to skeletal toxicity and growth plate changes such as steroids or endocrine replacement medications are warranted. This method has been incorporated into several clinical trials with angiogenesis inhibitors for children with refractory cancers or genetic tumor-predisposition syndromes such as NF1 or hereditary medullary thyroid carcinoma and clinical outcomes are pending.

The views expressed do not necessarily represent the views of the National Institutes of Health or the U.S. government. No potential conflicts of interest were disclosed.

We thank Jeffrey Baron, MD, National Institute of Child Health and Human Resources, for helpful discussions during data analysis and manuscript writing.

This study was supported by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.