Prevalence of Leukoencephalopathy in Children Treated for Acute Lymphoblastic Leukemia with High-Dose Methotrexate

Patient Population

Consecutive patients who were at least 1 year of age who were enrolled on our institutional ALL treatment protocol between July 30, 1998, and August 30, 1999, were eligible for this study. The diagnosis of ALL was established by means of morphologic, cytochemical, immunophenotyping, and genetic studies. Patients must have received no more than 1 week of previous therapy, which must have included only glucocorticoids. Because this study was designed to assess the longitudinal prevalence of LE in otherwise normal-appearing brain, patients with other neurologic complications, such as thrombosis, were not eligible. Written informed consent was obtained from the patient or his or her parent or guardian according to institutional review board, National Cancer Institute and Office for Human Research Protection guidelines. Fifty-three patients were enrolled on the treatment protocol. However, three patients were younger than 1 year, three died during induction, one withdrew from the protocol, and one had thrombosis at presentation. The exclusion criteria resulted in 45 patients eligible for this study: 21 male and 24 female patients aged 1.5–18.6 years (median, 5.4 years) at diagnosis (Table 3).

Treatment and Procedure Timeline

Patients were assigned to low-, standard-, and high-risk groups on the basis of comprehensive risk classification, which included blast cell immunophenotype and genotype, presenting clinical features, and early treatment responses. All patients received seven courses of IV MTX, with doses adjusted to achieve targeted systemic exposure to eliminate individual differences due to variability in clearance (19). The plasma steady-state concentration was 33 μmol/L in low-risk patients and 65 μmol/L in standard- or high-risk patients; average doses of MTX required to achieve these targeted concentrations were 2.5 and 5.0 g/m², respectively. Patients underwent MR examinations at four points during therapy: after one course (week 6, induction), after four courses (week 7, continuation), after seven courses (week 31, continuation), and at week 120 of continuation. Week 120 of continuation therapy was the end of therapy for the female patients, but the male patients received an additional 26 weeks of continuation therapy. The timing of this last imaging examination was chosen to ensure that all patients were evaluated at approximately the same point in therapy.

MR Imaging

LE is best visualized with a T2-weighted sequence, preferably with CSF attenuation (Fig 1). The imaging protocol was designed to simultaneously yield raw images for the segmentation procedures and images necessary for the clinical evaluation of the patients. All MR examinations were performed without contrast agent by using a 1.5-T Vision whole-body unit (Siemens Medical Systems, Iselin, NJ) with a standard circular polarized volume head coil. To minimize variability, a localizer sequence was used to determine the position of the patient in the coil. All images were acquired as 3-mm-thick contiguous oblique transverse-imaging sets defined by the most inferior extent of the genu and the splenium of the corpus callosum on the midline sagittal image. T1-weighted images were acquired with a multiecho inversion-recovery imaging sequence (TR/TE/TI/NEX = 8000/20/300/1, seven echoes). Proton density–and T2-weighted images were acquired simultaneously with a dual spin-echo sequence (TR/TE1/TE2/NEX = 3500/17/102/1). Fluid-attenuated inversion recovery (FLAIR) images were acquired with a multiecho sequence (TR/TE/TI/NEX = 9000/119/2470/1, seven echoes).

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

T1-, T2-, and proton density–weighted and FLAIR images (left to right) from a typical examination. LE can be seen in the white matter posterior to the ventricles.

Quantitative MR Detection of LE

Registration methods developed by Ostuni and colleagues (20) were used to register all MR imaging sets within an individual examination. A correction for radiofrequency inhomogeneities in the imaging sets was performed with a modified renormalization transformation combined with a local contrast ratio measure for in-section correction and with polynomial modeling for correction across the sections. A combined imaging set consisting of FLAIR and T1-, T2-, and proton density–weighted MR images, as well as white matter, gray matter, and CSF a priori maps from a spatially normalized atlas were analyzed with a neural network segmentation based on a Kohonen self-organizing map (21, 22). The self-organizing map essentially performed a pattern recognition task based on variations in the signal intensity properties of each point on the image. After the procedure was completed, the image was segmented into multiple tissue classes, with each one represented by a particular prototypical vector or pattern of intensity variation.

To improve the differentiation between true abnormal regions and CSF contaminated regions with high FLAIR signal intensity, an in-plane gradient magnitude threshold was added to the segmentation algorithm (23). Although CSF on the FLAIR images should be attenuated by the inversion pulse and appear dark, imperfect inversion and CSF pulsation can cause CSF at the extreme edges of the ventricles to sometimes remain bright, complicating the differentiation of true abnormal regions. After the segmentation was completed, the segmented maps were manually classified to identify the prototypical vectors associated with normal-appearing genu and diffuse white matter hyperintensities. The prototypical vector associated with the segmented region corresponding to areas of highest signal intensity on T2-weighted and FLAIR images were compared with normal-appearing genu vectors from the same patient. This enabled each patient to act as his or her own control and adjusted for known maturational changes in the relaxation properties of white matter at young ages and for acquisition variance caused by coil loading and tuning. This quantitative method was shown to be highly reproducible and to agree with radiologists’ reading of LE in 81% of cases (24).

Statistical Analysis

A Cox proportional regression was performed to estimate the effect of predictors on the development of LE. The first MR examination demonstrating development of LE was considered an event, and the number of weeks since diagnosis (corresponding to time in the treatment protocol) was the measure of time to event. Normal studies were censored, and the cumulative indices were then estimated. After the influential covariates (predictors) were identified, a generalized linear model with generalized estimating equations was applied to produce a logistic regression model that could predict the probability of LE in any patient treated in a similar protocol at any of the four time points in therapy. Changes in the modeled probability of LE between points in therapy were then tested for significance (alpha/0.05).