Apparent diffusion coefficient histogram analysis of neonatal hypoxic–ischemic encephalopathy
- First Online:
- Cite this article as:
- Cauley, K.A. & Filippi, C.G. Pediatr Radiol (2014) 44: 738. doi:10.1007/s00247-013-2864-1
Diffusion-weighted imaging is a valuable tool in the assessment of the neonatal brain, and changes in diffusion are seen in normal development as well as in pathological states such as hypoxic–ischemic encephalopathy (HIE). Various methods of quantitative assessment of diffusion values have been reported. Global ischemic injury occurring during the time of rapid developmental changes in brain myelination can complicate the imaging diagnosis of neonatal HIE.
To compare a quantitative method of histographic analysis of brain apparent coefficient (ADC) maps to the qualitative interpretation of routine brain MR imaging studies. We correlate changes in diffusion values with gestational age in radiographically normal neonates, and we investigate the sensitivity of the method as a quantitative measure of hypoxic–ischemic encephalopathy.
Materials and methods
We reviewed all brain MRI studies from the neonatal intensive care unit (NICU) at our university medical center over a 4-year period to identify cases that were radiographically normal (23 cases) and those with diffuse, global hypoxic–ischemic encephalopathy (12 cases). We histographically displayed ADC values of a single brain slice at the level of the basal ganglia and correlated peak (s-sDav) and lowest histogram values (s-sDlowest) with gestational age.
Normative s-sDav values correlated significantly with gestational age and declined linearly through the neonatal period (r2 = 0.477, P < 0.01). Six of 12 cases of known HIE demonstrated significantly lower s-sDav and s-sDlowest ADC values than were reflected in the normative distribution; several cases of HIE fell within a 95% confidence interval for normative studies, and one case demonstrated higher-than-normal s-sDav.
Single-slice histographic display of ADC values is a rapid and clinically feasible method of quantitative analysis of diffusion. In this study normative values derived from consecutive neonates without radiographic evidence of ischemic injury are correlated with gestational age, declining linearly throughout the perinatal period. This method does identify cases of HIE, though the overall sensitivity of the method is low.
KeywordsHypoxic–ischemic encephalopathyApparent diffusion coefficient mapsDiffusion-weighted imagingMagnetic resonance imagingBrain developmentNeonates
Diffuse hypoxic–ischemic injury in neonates can be a difficult radiologic diagnosis. MRI has become a clinical standard, but many factors contribute to difficulty in interpretation of these studies. First, such studies are infrequent except at the largest neonatal intensive care facilities, and most neuroradiologists see such studies only occasionally. Normal, healthy babies are rarely imaged, limiting the radiologist’s experience with neonatal brain MRI. Second, the neonatal brain undergoes rapid changes in myelination, and the appearance of the brain is a function of gestational age. Third, the diffuse nature of the injury can confound image interpretation. Although a number of imaging criteria for hypoxic–ischemic encephalopathy (HIE) have been proposed [1–3], objective criteria for making this difficult diagnosis are needed.
Diffusion-weighted imaging (DWI) is most sensitive for ischemia and stroke in adults but it can be difficult to interpret in neonates because incomplete myelination can result in a strong T2 shine-through effect, and for this reason ADC maps are recommended for the evaluation of neonatal hypoxic–ischemic brain injury . With the advantage of being quantitative, ADC maps have been investigated as a prognostic tool in neonatal HIE [4–7]. One major disadvantage of ADC map analysis is the need for region-of-interest placement, which introduces a subjective component to image analysis and yields a volume-averaged value, usually over a small area of the brain. More objective means of ADC analysis can increase the reproducibility of ADC measurements and improve the diagnostic and prognostic value of ADC data.
This study was undertaken to investigate the use of a clinically practical ADC histographic method in the evaluation of changes in neonatal brain diffusion in the perinatal period for evaluation of neonatal HIE. The aim of this study was to compare qualitative assessment of HIE as determined by two experienced neuroradiologists with quantitative values as determined by an ADC histogram.
Materials and methods
Normal studies. Values recorded from single-slice whole-brain region of interest (ROI) of radiographically normal neonates. Gestational age at birth (weeks, days) and age at scan time are given to yield gestational age at scan. Apgar scores are given at 1 and 5 min, with some studies including 10-min or 10- and 20-min scores. ROI areas (mm2) are given together with the average ROI value (cm2 × 10−2), with standard deviation for this measurement. ADC histogram peak (s-sDav), width of the peak at ½ maximum ADC value (2σ) and lowest ADC histogram value (s-sDlowest) are reported (cm2 × 10−2)
Gestational age at birth (weeks, days)
Age at scan (days)
Gestational age at scan (weeks)
ROI area (mm2)
Possible IVH on ultrasound
Midline facial defect
Seizure, possible abnormal US
Apnea, small EDH
Apnea, possible seizure
Hypoxic–ischemic encephalopathy (HIE) studies. Values recorded from single-slice whole-brain region-of-interest of neonates with radiographic signs of hypoxic–ischemic encephalopathy. Gestational age at birth (weeks, days) and age at scan time are given to yield gestational age at scan. Apgar scores are given at 1 and 5 min, with some studies including 10-min or 10- and 20-min scores. ROI areas (mm2) are given together with the average ROI value (cm2 × 10−5), with standard deviation for this measurement. ADC histogram peak (s-sDav), width of the peak at ½ maximum ADC value (2σ) and lowest ADC histogram value (s-sDlowest) are reported (cm2 × 10−2). Patient 11* demonstrated an abnormal bimodal histogram with peak values shown, s-sDav taken as 1,050 cm2 × 10−2. Neonates 2, 6, 7 and 9 died shortly after birth
Gestational age at birth (weeks, days)
Age at scan (days)
Gestational age at scan (weeks)
ROI area (mm2)
EEG, possible encephalopathy
Respiratory distress, seizure
US, possible HIE
Seizure, possible HIE
The studies were routine clinical studies and the interpreting neuroradiologist had access to the clinical records at the time of interpretation. A second review of the studies was performed by one of the authors (K.A.C.) in the review of the data for this study. The second reviewer was blinded to clinical outcome subsequent to the time of the scan; medical records following the patients’ discharge from the NICU were generally not complete or not readily available through the electronic medical records archiving system of this institution.
The interpreting radiologist holds a board certification from the American Board of Radiology (ABR) with Certificate of Advanced Qualification (CAQ) in neuroradiology and with more than 5 years of experience as an attending neuroradiologist. The second interpreter (K.A.C.) holds a board certification from the ABR with CAQ in neuroradiology and 3 years of experience as an attending neuroradiologist.
MRI data acquisition and analysis
Neonates underwent MRI as soon as practical after birth, as determined by the treating neonatal intensivists. All images were from a single 1.5-T scanner (Signa; GE Healthcare, Waukesha, WI) equipped with gradients with a maximum slew rate of 120 mT/m/ms and a maximum strength of 33 mT/m. A standard high-definition receive-only multi-element (eight elements) surface head coil was used. For imaging newborns were swaddled tightly, with heads restrained with cushions, while in the imaging system.
Axial images were acquired orthogonal to the anterior– posterior commissure line in standard fashion. The imaging protocol consisted of multiplanar T1-weighted imaging (axial: repetition time/echo time [TR/TE] 550/13 ms; sagittal: TR/TE 400/9 ms; coronal: TR/TE 12.8/5.37 ms), axial T2-weighted imaging (TR/TE 3,150/1,000 ms), axial fluid-attenuated inversion recovery (FLAIR) (TR/TE 902/142 ms), axial multiplanar gradient recalled echo (MPGR) (TR/TE 517/25 ms) and coronal spoiled gradient recalled echo (SPGR) (TR/TE 12.8/5.37 ms). Field of view was 170 × 170 mm or 180 × 180 mm for all series. Images were acquired at 4-mm slice with no gap, with the exception of the sagittal T1 sequence (3 mm, no gap) coronal T1 sequence (3.5 mm, 0.4-mm gap) and the coronal SPGR sequence (2 mm, 1-mm gap).
Axial DWI was acquired at three directions (TR/TE 10,000/94.4 ms, field of view 180 × 180 mm, 4-mm slice for 25 slices, no gap, number of excitations 2) with a b value of 800 s/mm2.
Single-slice ADC histogram generation and analysis
The s-sDav values were recorded as the peak value for each case. The value for s-sDlowest was taken at the leading edge of the histogram where voxel frequency was approximately 5 pixels. The tissue-compartment s-sDav distribution width (σ) of the histogram is determined at ½ peak value  and is also reported. The s-sDav and s-sDlowest histographic values of radiographically normal neonates were plotted as a function of the gestational age at the time of the scan. The s-sDav and s-sDlowest ADC values of radiographically normal neonates were also plotted together with neonates radiographically determined to have hypoxic–ischemic encephalopathy.
We performed statistical analysis including scatter plots with linear regression as well as Pearson correlation and confidence bands using statistical software (SPSS 20; SPSS, Chicago, IL). P < 0.05 was considered significant.
We generated single-slice ADC histograms and recorded peak (s-sDav) and lowest (s-sDlowest) ADC histographic values (Tables 1 and 2). We also evaluated single-slice ADC map histographic values of three normal adults to confirm that the single-slice ADC histogram and Dav values were comparable to those previously reported (750.0 +/− 30.0 × 10−2 cm2/s) .
In our analysis of ADC histograms, 6 of 12 radiographically determined cases of hypoxic–ischemic encephalopathy demonstrated peak histogram values that were significantly below the 95% confidence bands for radiographically normal neonates. One HIE study demonstrated higher than normal ADC signal, and five neonates with HIE demonstrated ADC values comparable to those of neonates without radiographic signs of HIE (Fig. 5). For lowest ADC value, an additional case fell outside the 95% confidence interval (Fig. 5). Four cases of HIE resulted in neonatal mortality, but there was no correlation between these cases and the degree of diffusion abnormality measured at the single time point we recorded. With the assumption that hypoxia or ischemia is most likely to occur at the time of birth, diffusion values in neonates with hypoxic–ischemic encephalopathy were compared with the time interval between birth and the time of MRI scan, though no significant correlation was found.
Neonatal hypoxic–ischemic encephalopathy (HIE) can be a difficult imaging diagnosis, and a number of imaging criteria have been proposed [3, 7, 10]. Recent findings that therapeutic hypothermia can improve neurological outcomes [10–12] together with the promise of other medical therapies to ameliorate ischemic damage in development [13, 14] have served to further emphasize the need for rapid and accurate identification of infants with HIE. ADC maps have been advocated as being sensitive in making the diagnosis of HIE, and the quantitative nature of ADC maps has led to a number of studies detailing quantitative use of ADC maps as prognostic tools for analysis of HIE [4–7]. The advantage of the quantitative capability of ADC measurements is tempered by the need for region-of-interest (ROI) analysis, which necessitates operator placement of ROI on images lacking in anatomical detail, and ROI measurements represent volume averaging over all the voxel values included in the ROI.
Though experienced pediatric neuroradiologists are likely to agree on interpretation of pediatric brain MRIs where a hypoxic–ischemic insult has occurred, interpretation remains subjective and difficult for less-experienced personnel. Our study was designed to compare subjective interpretation by two experienced neuroradiologists with the more objective histographic analysis of diffusion maps. The goal of our study was to introduce this method of diffusion data analysis to the clinician and to test the concordance of this method with the interpretation by two neuroradiologists. Concordance with follow-up data would be an important direction for future study, to be pursued only if the methods we present are sufficiently promising.
ADC values can be presented as a histographic display of frequency of voxel occurrence as a function of ADC value and offer information not evident on traditional ROI analysis, such as peak ADC value, lowest ADC value and overall shape of the ADC histogram. For these reasons ADC histographic analysis has been advocated in a number of other studies as a more reproducible method of ADC map interpretation [9, 15, 16]. Published studies of ADC histographic methods in the brain entail whole-brain analysis and use proprietary software that is not generally available.
A compromise approach between ROI methods and whole-brain histographic display can be achieved by histographic display of ROI data. The GE FuncTool enables rapid analysis of single-slice ADC maps to generate a representative pixel-by-pixel histographic display of ADC values. Using this method ROI data from an entire brain slice can be rapidly displayed without data manipulation or post-processing and readily used in the clinical setting. We investigated the use of this method in the evaluation the diffusion properties of neonatal brain as a function of gestational age and in the evaluation of diffuse hypoxic–ischemic encephalopathy.
Single-slice ADC histograms of neonates without radiographic signs of hypoxic–ischemic encephalopathy show a Gaussian ADC pixel distribution similar to ADC histograms previously described for the whole brain in normal children and adults (Fig. 1) [9, 15, 16]. The recorded values are comparable to those noted using ROI approaches [4, 7, 17, 18] or whole-brain histograms [15, 16] of neonates of similar gestational ages. Single-slice ADC histographic metrics of s-sDav and s-sDlowest showed a strong linear correlation with gestational age, and Dav values fell within a narrow range of values around the regression line.
A previous study investigated the changes in diffusion in gray and white matter of perinatal brain as a function of gestational age. In a prospective study of 22 normal newborns using diffusion tensor methods and an ROI method of analysis, Neil et al.  showed a linear decline in regional brain diffusion measurements as a function of gestational age. That study showed a linear decline in white matter diffusion from 1.7 × 10−3 mm2/s at 32 weeks to 1.3 × 10−3 mm2/s at 42 weeks, and gray matter of 1.3 × 10−3 mm2/s at 32 weeks to 1.2 mm2/s at 42 weeks. Our single-slice ADC values derived from a histographic display of routinely acquired diffusion data of radiographically normal (but clinically variable) neonates yielded values very similar to those reported in the earlier study, with peak value of 1.4 × 10−3 at 34 weeks and 1.2 × 10−3 at 42 weeks. The data spread (r = 0.69) appears to be less for the histographic method, although r values were not reported for the earlier study.
Another study of perinatal brain injury was designed to determine the magnitude and time course of the changes in water diffusion (Dav) following a clear-cut event near birth that could be timed accurately and was likely to be the cause of the brain injury . Using ROI methods of analysis, these investigators found that Dav values were decreased in most infants 1 day after injury, but injury was not evident or was underestimated in 4 of 10 infants despite the presence of injury on conventional imaging at 1 week of age. Decreased Dav values were most evident 2–7 days after the injury, but by the 7th day Dav values were returning to normal (pseudonormalization), and appeared to be higher than normal after 7 days . This suggests that the sensitivity of MR diffusion methods is dependent on the timing of the study relative to the time of the hypoxic–ischemic insult, which is not always known.
Though absolute measures of ADC are not always sensitive because of the Gaussian distribution of normal ADC values (Fig. 1), s-sDlowest data suggest that any ADC ROI value below 800 cm/s2 and measured within a single slice at the level of the basal ganglia is abnormal (Figs. 2 and 5). Further, the normative values curves (Figs. 2 and 5) show that gestational age can be important for the interpretation of ADC values because there is a rapid decline in Dav through the perinatal period (from regressed value of 1,400 × 10−2 cm/s2 at 34 weeks to 1,200 × 10−2 cm/s2 at 42 weeks, or 14.2%). The scatter plot shows, for example, that at 40 weeks a Dav of 1,150 × 10−2 cm/s2 would fall into the normal range, where this value would not be normal at 36 or even 38 weeks. Greater water content of the brain as reflected in higher ADC values might reflect a relative insensitivity of ADC for infarct at earlier gestational ages. In our cohort all cases of significantly decreased ADC values fell within weeks 37–41 of gestational age. Varying the parameters of diffusion acquisition (such as b values) or post-processing (such as analysis of eADC values) might mitigate pseudonormalization effects and add sensitivity to this method of analysis. Larger cohort size would permit increased statistical power to these claims. Although quantitative ADC mapping using histographic methods is not necessarily sensitive because of pseudonormalization effects, it might be helpful to the diagnostic radiologist in the interpretation of neonatal brain ADC maps.
Study limitations and future directions
This is a pilot study to evaluate the use of ADC histographic analysis as a tool for aiding in making the diagnosis of neonatal HIE. This study is limited to imaging findings and attempts to correlate qualitative imaging findings and the impression of two experienced neuroradiologists with the quantitative findings derived from ADC histograms. Healthy neonates are not typically imaged, and there are very few studies reporting the ADC values of healthy newborns [19, 21]. All the neonates reported in this study were imaged because of signs and symptoms thought to be clinically significant for neurological finding and therefore were not normal as evidenced by the Apgar scores (Table 1). Our normal cohort therefore included studies without radiographic evidence of abnormality as determined by two neuroradiologists. Studies identified in this way showed a strong correlation with gestational age using linear regression, with few outliers (Fig. 2), so the assumption of normalcy appears to be valid with respect to diffusion, though slight diffusion abnormalities might be present and decreased scatter and tighter correlation with gestational age might be expected with studies of healthy newborns. Normative MRI data on neonatal brains is of high interest to diagnostic radiology and clinical research, but MRI study of truly normal premature infants for the sole purpose of obtaining normative data cannot be justified. For this reason we pursued our study using the data that were available to us with the belief that these data, and data from other similar studies, would serve to approximate the normative data set that is unobtainable. Additionally, this comparative data set is inclusive of NICU babies with the spectrum of neonatal comorbidities characteristic of this population and so should not be readily discounted as an inappropriate cohort to compare with NICU babies with hypoxic–ischemic encephalopathy. The comparative data set was acquired in the same retrospective fashion, over the same time frame and from the same MR scanner as the neonates with presumptive HIE, minimizing concerns over possible differences in patient preparation or differences in scanners and software calibration between the study cohorts, and minimizing the effects of other potentially confounding variables.
Our study is a retrospective review of clinical data. The timing of the insult thought to be causal for HIE is not typically known. Infants are imaged as soon as deemed practical after birth. In this environment the timing of the scan may be suboptimal for evaluation of HIE, and the effects of pseudonormalization of ADC values cannot be controlled.
An easy solution to the problem of pseudonormalization would be to standardize the timing of the scan from the time of the insult. But this is usually not possible because the time of the insult is not always known and the infant might not be stable for scanning at a particular time point after the presumed insult. As such, future studies of brain diffusion in HIE are needed to address the problem of pseudonormalization through other strategies such as use of a higher magnetic field strength (3T), optimization of scan parameters including b values, scanning at multiple time points, and use of diffusion tensor imaging.
Hypothermia and head cooling are a promising new therapy for HIE, and the effect of therapeutic hypothermia and head cooling on HIE outcomes, and on the correlating imaging findings and specifically on diffusion values, is a very important topic that might be addressed in animal models but could also be addressed in controlled clinical studies.
There is also increasing interest in the standardization of quantitative MRI values such that standard curves of cases such as the one we derived here might be used to establish standard deviations and would be broadly useful to other institutions. Until that is achieved, however, institutions need to generate a limited amount of data such as we describe to establish the accuracy and reproducibility of the diffusion numbers they obtain.
Finally, the number of neonates in this study with radiologic findings of global and diffuse HIE is relatively small (12) but sufficient to illustrate the strengths and limitations of the technique we present. Future studies and studies of larger cohorts would serve to improve the statistical power of this study’s findings.
Single-slice histographic analysis of ADC values is a rapid and clinically useful technique for estimation of Dav. Regression analysis of single-slice Dav values of neonates obtained in this way is significantly correlated with gestational age during the neonatal period and the findings are consistent with reported values using more involved methods. The normative values distribution aids in the identification of HIE. Six of 12 cases of HIE could be identified with ADC histographic analysis, although in some cases there might have been a pseudonormalization of diffusion restriction. Histographic analysis of ADC maps might prove to be a useful tool for the evaluation of neonatal hypoxic–ischemic injury, though other radiographic signs of HIE remain essential in making this sometimes difficult diagnosis.
Conflict of interest