Quantification of Diffusion Magnetic Resonance Imaging for Prognostic Prediction of Neonatal Hypoxic-Ischemic Encephalopathy

Abstract Neonatal hypoxic-ischemic encephalopathy (HIE) is the leading cause of acquired neonatal brain injury with the risk of developing serious neurological sequelae and death. An accurate and robust prediction of short- and long-term outcomes may provide clinicians and families with fundamental evidence for their decision-making, the design of treatment strategies, and the discussion of developmental intervention plans after discharge. Diffusion tensor imaging (DTI) is one of the most powerful neuroimaging tools with which to predict the prognosis of neonatal HIE by providing microscopic features that cannot be assessed by conventional magnetic resonance imaging (MRI). DTI provides various scalar measures that represent the properties of the tissue, such as fractional anisotropy (FA) and mean diffusivity (MD). Since the characteristics of the diffusion of water molecules represented by these measures are affected by the microscopic cellular and extracellular environment, such as the orientation of structural components and cell density, they are often used to study the normal developmental trajectory of the brain and as indicators of various tissue damage, including HIE-related pathologies, such as cytotoxic edema, vascular edema, inflammation, cell death, and Wallerian degeneration. Previous studies have demonstrated widespread alteration in DTI measurements in severe cases of HIE and more localized changes in neonates with mild-to-moderate HIE. In an attempt to establish cutoff values to predict the occurrence of neurological sequelae, MD and FA measurements in the corpus callosum, thalamus, basal ganglia, corticospinal tract, and frontal white matter have proven to have an excellent ability to predict severe neurological outcomes. In addition, a recent study has suggested that a data-driven, unbiased approach using machine learning techniques on features obtained from whole-brain image quantification may accurately predict the prognosis of HIE, including for mild-to-moderate cases. Further efforts are needed to overcome current challenges, such as MRI infrastructure, diffusion modeling methods, and data harmonization for clinical application. In addition, external validation of predictive models is essential for clinical application of DTI to prognostication.


Introduction
Neonatal hypoxic-ischemic encephalopathy (HIE) is the most common cause of neonatal encephalopathy and one of the major causes of neonatal acquired brain injury.HIE is caused by the disruption of cerebral blood flow and oxygen supply during the perinatal period [1].Neonatal HIE may lead to significant long-term neurological disorders or death, and its incidence rate is approximately 1.5 cases per 1,000 total live births [2].Although therapeutic hypothermia (TH) has reduced mortality and morbidity by one-third and has become a standard treatment [3][4][5][6][7], a significant proportion of survivors still develop residual neurological outcomes.However, identifying those babies at risk remains a significant challenge.An accurate and robust prediction of short-and longterm outcomes may provide clinicians and families with fundamental evidence for their decision-making, the design of treatment strategies, and the discussion of developmental intervention plans after discharge [8].Therefore, the development of accurate prognostic indicators has been the focus of recent research in the field of neonatal HIE.
However, one of the biggest challenges in prognostication using conventional anatomical MRI (e.g., T1-and T2-weighted images) is the low sensitivity.HIE children with severe neurological outcomes may lack significant MRI findings during neonatal period (first 28 days of life) [19,[27][28][29][30].To overcome this challenge, various types of image quantification methodologies have been introduced and applied to quantitative MRI modalities, such as diffusion MRI [31,32] and MR spectroscopy [28,[33][34][35].Diffusion MRI quantifies the thermal motion of the water molecules in the tissue and uses it as a probe to detect microstructural alterations that cannot be visualized by anatomical MRI [36][37][38].MR spectroscopy utilizes the chemical properties of the tissue to probe metabolic alterations in the brain.
This review focuses on the prognostic value of diffusion MRI, particularly diffusion tensor imaging (DTI).An overview of study designs and image quantification methods, including hypothesis-driven approaches based on quantification of specific regions of interest (ROI) [39] to data-driven approaches, such as voxel-based analysis using tract-based spatial statistics (TBSS) [40], and atlasbased methods for image quantification (ABA), is provided.Studies that examine how DTI quantification values are changed by HIE and the correlation between these changes and HIE severity will be introduced.Methods with which to set cutoff values for DTI quantification values, and thus, identify severe cases and prognostic models based on whole-brain quantification approaches combined with machine learning, will also be presented, along with the ability of these methods to predict the prognosis of HIE patients.

Basic Principles of DTI and Quantification
Diffusion MRI quantifies the thermal motion of water molecules in tissues using mathematical models, from which microscopic anatomical structures can be characterized [36][37][38].The simplest model is the average of the magnitude of diffusion in three orthogonal directions, called a mean apparent diffusion coefficient (ADC) map, which is widely used in clinical practice to diagnose acute ischemic events in the brain, including neonatal HIE.A mean ADC map can detect acute to subacute HIE lesions that are normal or only slightly abnormal on conventional MRI [19,[27][28][29][30].However, the information obtained from the mean ADC map alone does not provide a detailed description of the various pathological processes seen in HIE.
DTI introduces a second-order tensor model that fits the diffusion of water in each voxel to an ellipsoid.Compared to the simple average model that provides the mean ADC map, DTI enables the generation of various measures that represent more detailed properties of the tissue, such as the size, shape, and direction of water diffusion that can be calculated from the tensor.Widely used scalar measures are fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity, and radial diffusivity, which can be calculated from the first, second, and third eigenvalues (λ 1 , λ 2 , λ 3 ) of a tensor [41][42][43][44] (Fig. 1).Since the characteristics of the diffusion of water molecules in tissues are affected by histological factors, such as the density of macromolecules, the orientation of structural components, and cell density, these parameters have often been used to study the normal developmental trajectory of the brain, particularly the white matter.
Maturation of cerebral white matter is characterized by the formation of dense and thick fiber bundles and myelin sheaths, both of which are actively ongoing processes during the perinatal period [45][46][47].The formation of tightly packed fiber bundles and the myelination both result in reduced extracellular space and restriction of water diffusion toward the direction perpendicular to the direction of fiber bundles [48,49].MD is the mean of the three diffusion tensor eigenvalues and represents the magnitude of diffusion (Fig. 1).When the extracellular space is reduced due to white matter maturation, MD decreases because water diffusion is more restricted.FA is a measure of the degree of anisotropy of diffusion and takes a scalar value between 0 and 1. Zero means that diffusion is isotropic, i.e., there is no preference for a particular direction, and 1 means that diffusion is completely anisotropic, i.e., it occurs only in a particular direction (Fig. 1).FA increases as fiber bundles pack more tightly during white matter maturation, reducing diffusivity in the direction perpendicular to the white matter tracts.
Another major feature of DTI is the delineation of white matter fiber bundles as streamlines [50][51][52].Since water molecules tend to diffuse in the direction of the fiber bundles, quantifying the orientation of the diffusion distribution of water molecules from diffusion MRI data can provide indirect information about how the white matter fiber bundles travel.The threedimensional modeling technique used to create these streamlines is called tractography, and the streamlines created are called tractograms.During the course of brain development, fiber density and the diameter of fiber bundles increase.Therefore, the types of white matter fiber bundles for which tractograms can be delineated increase [53].
The various scalar maps produced from DTI can be used for statistical analysis, such as group comparisons, or as input variables for creating prognostic models.For this purpose, scalar values derived from the tensor (e.g., MD and FA) must be quantified for each anatomical region.Commonly used methods for this purpose include ROIbased analysis, TBSS, and ABA [54].
For study designs based on hypotheses, such as that certain brain regions are more likely to be impaired by HIE, ROIs are often used to quantify DTI.In this case, preselected ROIs are drawn in the specific regions based on the hypotheses, and the average scalar value of pixels in the ROI is calculated.ROI drawing can be done manually or by automatic image parcellation methods.When a specific fiber tract defines the ROI, fiber tractography is used as a means of delineating the ROI.The term "tract-based analysis" is commonly used to describe this type of DTI quantification method.
To search brain regions affected by HIE in a datadriven approach, rather than being based on a specific hypothesis, DTI is quantified for the whole brain.TBSS is a method with which to quantify scalar values on a voxelby-voxel basis for entire white matter regions based on skeletonization of group-averaged FA maps [55,56].This method is often used for the voxel-by-voxel statistical analysis as described in the following sections.
ABA is based on whole-brain image parcellation.Each DTI is divided into multiple ROIs that cover the whole brain, allowing quantification of DTI-derived scalar values for each ROI that defines a specific anatomical structure or region (Fig. 2).Since whole-brain image parcellation is prohibitively difficult to perform manually, various automated methods have been proposed [57][58][59][60]; once DTI is parcellated, whole-brain ROI-by-ROI statistical analysis can be performed.A limited number of ROIs can also be selected from the set of all ROIs, based on a hypothesis, and used for ROI analysis for hypothesis testing.
TBSS and ABA can be used in combination: after image quantification of the FA skeleton on a voxel-byvoxel basis, anatomical parcellation maps obtained with the atlas-based approach can be overlaid on the FA skeleton to segment the FA skeleton and perform image quantification based on the skeletons-of-interest.It is also possible to annotate the skeleton based on the anatomical labels of the reference atlas [61,62].
DTI has been instrumental in elucidating the developmental patterns of brain white matter fibers.In principle, brain maturation proceeds from the most ventrocaudal parts of the brain toward the most dorso-rostral parts.That is, maturation begins from the brainstem toward the cerebrum [47,54].Within the cerebrum, the direction of maturation is from the center to the periphery and from posterior to anterior [63][64][65].Among the centrally located fiber bundles, the corticospinal tracts (CSTs) are the earliest to mature [66].Within the cerebral cortex, the primary cortex matures ahead of the association cortex [65,67].Understanding the relationship between structure-specific maturation patterns of white matter bundles and the changes observed in DTI measurements (especially FA and MD) is fundamental to the interpretation of DTI results seen in pathological conditions, including neonatal HIE.
There is a caveat to the use of DTI in research: because DTI measurements are very sensitive to differences in scanner types and scan parameters, it is preferable to use a single scanner and fixed scan protocol in scientific studies.When analysis is based on DTI scanned with various scanners and scan parameters, as is often the case in multicenter studies, mathematical models need to be introduced to harmonize the data [68][69][70].In general, the larger the voxel size, the lower the FA value.This is due to the loss of diffusion anisotropy as voxel size increases, as it is more likely to contain multiple fiber structures oriented differently within each voxel.

Neonatal HIE and DTI Changes
The main pathogenesis of HIE can be roughly divided into primary (0-6 h after initial injury) and secondary (6-48 h after initial injury) periods of energy failure [1,71].During primary energy failure, insufficient cerebral blood flow and oxygen supply cause acute dysfunction of mitochondria and transition from aerobic to anaerobic glycolysis, leading to ATP deficit and the disruption of the Na+/K+ ATPase pump.These events result in cytotoxic edema, which is a redistribution of water from the extracellular to the intracellular space due to the osmotic stress [1,72].In DTI, cytotoxic edema is expressed by a decrease in MD and ADC, reflecting a decrease in extracellular space that restricts the available space for free water molecules to diffuse [73].In white matter regions, an FA increase can be observed in the acute phase of brain injury because diffusion in the direction orthogonal to the fiber bundle is more reduced than diffusion in the direction parallel to the fiber bundle [74] (Fig. 3).
During secondary energy failure, free radical release due to the reperfusion and reoxygenation of cerebral tissues causes oxidation stress.This stress damages the blood-brain barrier function, which is followed by increased vascular permeability, and causes extravascular leakage of serum proteins.Because of the low capacity to remove free radicals, this leakage induces prolonged extracellular fluid retention and results in vasogenic edema [30].In DTI, vasogenic edema is expressed by increased MD, reflecting increased extracellular fluid retention, thus allowing water molecules to diffuse more freely with less restriction [75].In white matter regions, FA tends to decrease with increasing MD because of relatively free diffusion in the extracellular space [76] (Fig. 3).
Hypoxic-ischemic conditions cause neuronal death, followed by Wallerian degeneration, a retrograde degeneration of the distal end of the axon [77,78].Hours to 2 weeks after neuronal death, axonal and myelin debris accumulate in lesions where Wallerian degeneration has occurred.This restricts water diffusion, resulting in a marked decrease in FA and MD values [79,80].Wallerian degeneration is typically seen throughout the CST, including the posterior limb of the internal capsule (PLIC) and the cerebral peduncle [81] (Fig. 3, 4).In the chronic phase, neuronal loss and gliosis cause a decrease in FA and an increase in MD (Fig. 3).
The results of tractography are also influenced by pathological changes related to HIE.If a major fiber tract, such as the CST, is damaged by HIE, the tractograms of that fiber may be disrupted, reduced in number, or lost [82].
In HIE neonates, the mixed effects of cytotoxic and vasogenic edemas, cell death, and Wallerian degeneration can be observed [77,83].According to the dominant effect of the pathogenesis, the DTI findings change chronologically (Fig. 5).In general, over the first week of HIE, FA values may increase (Fig. 6) or decrease (Fig. 7), while MD and ADC values decrease (Fig. 6, 7).The decreased MD and ADC values return to the normal range in the second week, a finding termed pseudo-normalization [84][85][86].Thereafter, MD and ADC values increase over the chronic phase, but FA values decrease [84,86] (Fig. 8).As discussed in the following sections, the degree of change and the location where the change is seen depend on the severity of the HIE.
The first step for predicting neurological outcomes is to identify the anatomical structures affected by HIE and to understand how DTI measurements are altered in these structures.Toward this goal, group comparisons have been made between HIE and control groups.In a  cross-sectional study, an ROI-based study performed on DTI acquired within 1 week after birth identified decreased MD in the putamen, thalamus, anterior and posterior limbs of internal capsule, occipital white matter, and the CST, compared to age-matched healthy infants [87].A study targeting DTIs acquired within 15 days after birth and focused on the metencephalon and thalamus demonstrated that decreased MD in the superior cerebellar peduncle and decreased FA in the middle cerebellar peduncle were seen in the severe HIE group compared to the control group [88].Another study that included even older neonates scanned within 3 weeks after birth demonstrated increased ADC accompanied by decreased FA in the HIE group compared to the non-HIE control group, in the basal ganglia (BG), thalamus, PLIC, cerebral peduncle, CST, and peripheral white matter areas [89], and increased MD accompanied by decreased FA in the PLIC, superior corona radiata, corpus callosum (CC), and external capsule [90].The discrepancy between decreased and increased diffusivity between these studies probably reflects the timing of the scan; the former study might have mainly captured the effects of cytotoxic edema, while the latter mainly captured vascular edema and cell death.These studies highlight the importance of the timing of the scan in interpreting the findings.
While ROI-based studies could delineate which DTI measures are affected in which anatomical structures, only a limited number of preselected brain areas have been explored; huge brain areas remain unexplored.To overcome the limitation of ROI-based analysis, TBSS was used to explore the whole white matter skeleton in a voxel-by-voxel manner.According to the group comparisons between HIE and healthy controls that targeted DTIs acquired at 1-4 weeks [90][91][92], decreased FA [91,92] and increased MD [90] and radial diffusivity (mean of the second and third eigenvalues) were observed and were widespread, especially in patients with HIE who had not received TH [91].The areas of impairment were significantly smaller for the group of HIE neonates who received TH compared to those without TH, suggesting the neuroprotective effect of TH and importance of subgrouping HIE neonates by the use of TH.
The next issue is whether DTI findings are associated with the severity of the neurological outcome.In general, severe HIE has been proven to be associated with more severe and extensive gray matter and white matter alterations compared to neonates with moderate or mild HIE.
On DTI images obtained within 6 days of birth, HIE neonates with low FA in the PLIC and cerebral peduncle and low ADC in the PLIC and BG were demonstrated to have poor neurological function according to the Amiel-Tison neurological assessment [93].A reduction of FA was observed in moderate-to-severe HIE for the first 3 weeks of life, but ADC was reduced in the white matter,  the BG, and the thalamus in only some severe HIE newborns, with pseudo-normalization at 2 weeks [86].This suggests that FA is a sensitive marker for the detection of anatomical changes associated with HIE, and ADC is a marker for differentiating severe cases from moderate HIE.
ABA performed with diffusion MRI showed that severe HIE neonates scanned at approximately 1 week of age [94] showed an extensive reduction in FA, while moderate HIE neonates showed a reduction in FA only in the inferior association fibers (inferior longitudinal fasciculus and inferior fronto-occipital fasciculus).In addition, diffusion MRI measurements obtained by nontensor modeling, such as fiber density, fiber crosssection, and fiber density and cross-section obtained by pixel-based analysis, demonstrate the potential for highly sensitive detection of anatomical changes seen in mild-to-moderate HIE newborns [94].The topic of nontensor modeling is discussed in the section "Remaining Challenges and Future Directions."

DTI for Neuroprognostication in HIE
Attempts have been made to predict the severity of neurological symptoms from DTI measurements in HIE neonates.Such studies are categorized as those aimed at predicting neurological status at the time of scanning (state prediction), around 1 month (short-term prognosis), and after 1 year of age (long-term prognosis) [95].
In a study aimed at predicting the neurological state during the first 2 weeks of life, FA values quantified at the anterior limb of the internal capsule using the JHU_MNI atlas [96] could predict good neurological status, defined by a neonatal behavioral neural assessment score [97] of 35 or higher, with a sensitivity of 70% and specificity of 80%, using FA ≥0.395 as the cutoff value [98].Note that this cutoff FA value may not necessarily be generalizable since FA values can take different values depending on scan parameters, as described in the section "Basic Principles of DTI and Quantification." In an early attempt to use FA values obtained from thalamic ROIs at 2 weeks of age to predict short-term prognosis, FA values were lower in HIE than in controls but did not correlate with length of hospital stay, time to oral intake, or presence of seizures [39].However, a recent study combined atlas-based image quantification with a least absolute shrinkage and selection operator regression model to detect changes in DTI, i.e., FA and MD changes in limbic, frontotemporal, and corticospinal projection fibers, even in mild HIE.The score obtained from the least absolute shrinkage and selection operator regression model, named the composite DTI (cDTI) score, correlated strongly with short-term neurological outcome (rho = 0.83) and was also associated with changes in serum biomarkers, such as IL10 and tau [99].Notably, although voxel size was controlled, the study included four different scanners and slightly different scanning protocols, but there was no significant effect of scanner and scanning parameters on the cDTI score.This study demonstrates the potential of wholebrain, data-driven approaches in quantifying lesions related to neurological outcomes and the potential use of DTI for neurological prognostication in a multicenter setting.
Since fine motor and language skills are difficult to assess in infancy, it would be important to evaluate whether DTI acquired in the neonatal period can predict neurological prognosis after the age of 1 year, when these neurological functions can be assessed [95,100].A tractsof-interest study of CC and CST showed an association between lower FA and lower mental and psychomotor developmental indices assessed at 15 and 21 months [101].In a study [40] using the TBSS to identify white matter skeletons associated with neurological ability at 12-28 months of age based on the Griffiths Mental Development Scales-Revised, FA values in the most significant voxels of the cerebral peduncle correlated well with developmental quotient (R 2 = 0.42), locomotor score (R 2 = 0.28), and eye-hand coordination (R 2 = 0.31).FA values in the CC showed good correlations with the personal-social score (R 2 = 0.33) and the hearing and language score (R 2 = 0.3) and FA values in the PLIC with the performance score (R 2 = 0.37).
Several attempts have been made to establish cutoff values with which to predict the occurrence of neurological sequelae that become evident after 1 year.MD and FA assessed at 4-16 days of age could be used as predictive markers for poor prognosis assessed at 1 year of age, i.e., death, cerebral palsy, dysarthria, and general developmental delay (general developmental quotient less than 88.7) as measured by the Griffith scale [102].In this study, receiver operating characteristic curve analysis was introduced to establish cutoff MD and FA values with which to predict poor prognosis.MD measured in the CC, thalamus, caudate, and frontal white matter had an area under the curve (AUC) of 0.72-0.78,a positive predictive value of 1, and a negative predictive value of 0.83, while FA in the frontal white matter had an AUC of 0.94, a positive predictive value of 0.71, and a negative predictive value of 1, indicating a reliable predictability of poor outcome.A recent study has largely replicated the finding: the FA value measured at the PLIC, centrum semiovale, and cerebral peduncle from images scanned at 10-14 days of age could predict the occurrence of cerebral palsy at 1 year of age [103].In particular, PLIC with a cutoff FA value of ≤0.435 exhibited an AUC of 1.0 (sensitivity 1.0, specificity 1.0) and the centrum semiovale with a cutoff FA value of ≤0.235 exhibited an AUC of 0.95 (sensitivity 0.94, specificity of 0.94), showing excellent predictive ability.
Rather than establishing a specific cutoff value for prediction, an attempt has been made to introduce a classification algorithm: a multidimensional marker derived from tractography using the non-tensor method was introduced to predict the presence of abnormal muscle tone on neurological examination at 2 years of age [104].This prediction was based on a random forest classifier that used 19 pairwise connections with abnormal tractography findings as input values, with an excellent classification performance of 0.99 sensitivity and 1.0 specificity to identify a group of neonates who developed abnormal muscle tone at 2 years of age.
These studies demonstrate that neonatal DTI has excellent potential to accurately predict both shortand long-term severe neurological outcomes.However, the number of HIE neonates included in each study was less than 60, and none of the prediction models have been validated in an external cohort of neonatal HIE; multicenter validation studies with larger neonatal HIE cohorts are essential for the clinical application of DTI as a prognostic predictor.In addition, there are no studies of milder neurological sequelae that become evident after school age, such as neuropsychiatric symptoms and learning disabilities; further research is needed to determine whether DTI can predict longer term outcomes.

Remaining Challenges and Future Directions
The purpose of prognostication is to provide clues for medical decision-making.Although DTI has shown potential as an accurate prognostic tool, there are still issues that need to be resolved before its clinical application.
First, performing a MRI scan on vulnerable neonates is not always easy.Especially for neonates who are fitted with infusion lines and monitors, transfer to the scanner room requires the assistance of multiple staff members.The noise and vibration caused by the gradient applied during the scan arouses the neonate, causing motion artifacts and reducing the success rate of the scan.Hardware improvements, including improvements to hospital facilities, may be needed to solve these problems.For example, some advanced neonatal intensive care units have solved these problems by installing an MRI scanner in the neonatal intensive care unit [105,106].Attempts are also being made to improve the success rate of scans by reducing motion artifacts and scan time [107][108][109].Such efforts can also improve the quality of the studies.The quality of neonatal DTI studies is compromised by the small number of participants; improving the success rate of DTI scans would allow better studies.
Second, some of the studies presented in the previous section used non-tensor models to quantify the effects of HIE on the brain [90,104], suggesting the possibility of detecting minor anatomical abnormalities with higher sensitivity than conventional DTI measurements.Conventional tensor models assume that the diffusion follows a three-dimensional Gaussian distribution and estimate one tensor field within each voxel.While this method is effective for extracting gross features of diffusion, such approximations are considered oversimplified when more detailed features are needed [110].Non-tensor models were introduced to describe non-Gaussian diffusion and to resolve fiber orientation within voxels that cannot be quantified by tensor models [111][112][113].Further studies are needed to determine whether non-tensor models such as fixel-based analysis and non-tensor-based probabilistic tractography can provide better prognostic prediction than a conventional tensor model.Predictive modeling by applying machine learning to input variables from the whole brain has also been introduced in several studies [99,104], showing the potential to maximize the prognostic predictive ability of DTI by comprehensively identifying imaging changes that may contribute to the pathogenesis of HIE.Such an approach is expected to improve the sensitivity of detecting minor pathological changes, as demonstrated in [99].
Third, standardized methods to harmonize DTI data have not yet been established.Most studies, to date, have been based on small numbers of subjects, resulting in the risk of selection bias and overfitting of predictive models to the study population.Predictive models with a larger number of subjects are needed for clinical application, thus necessitating multicenter studies and data sharing.However, DTI measurements are sensitive to differences in scanner and scan parameters, hindering the use of technically heterogeneous DTI.Several attempts have been made to harmonize DTI data [68][69][70], and such efforts are essential for future multicenter studies and clinical applications.

Conclusion
DTI is one of the most powerful neuroimaging tools for the prediction of neonatal HIE prognosis by providing microscopic features that cannot be assessed by conventional MRI.Applying a data-driven, unbiased approach using machine learning techniques to these features obtained by whole-brain imaging quantification could provide more accurate predictions of HIE prognosis than those based on DTI quantitative values obtained from a single ROI.For clinical application, further efforts are needed to overcome current challenges, such as MRI infrastructure, diffusion modeling methods, and data harmonization.In addition, external validation of the prediction model is essential for the clinical application of DTI for prognostic prediction.

Fig. 1 .
Fig. 1.Schematic representation of the relationship between water diffusion, tensor fitting, and scalar values obtained from the tensor.The water molecule (red circle) and diffusion trajectory (red line) in the left figure are approximated by an ellipsoid (pink ellipsoid surrounded by a red dotted line) through tensor fitting.The three orthogonal axes of the ellipsoid are called the first, second, and third axes, and the length of each axis is called the first eigenvalue (λ 1 ), second eigenvalue (λ 2 ), and third eigenvalue (λ 3 ; λ 1 ≥λ 2 ≥λ 3 ).Various scalar measures can be obtained from the three eigenvalues.Fractional anisotropy (FA) and mean diffusivity (MD) are often used to estimate the tissue microstructure of the brain.

Fig. 2 .
Fig. 2.An example of a multi-contrast neonatal brain atlas (top) and atlas-based DTI parcellation (bottom).Lesions seen in the mean diffusivity (MD) map are highlighted in red (bottom middle).By deforming the neonatal atlas to fit the patient's brain and superimposing the deformed atlas parcellation map on the MD map (bottom right), MD values can be quantified for each anatomic region defined in the parcellation map.

Fig. 3 .
Fig. 3. Schematic representation of the pathological changes seen in neonatal HIE.The upper panel shows gray matter areas, and the lower panel shows white matter areas where myelinated axons form densely packed fiber bundles.Neurons are color coded in blue, myelin sheaths in light blue, astrocytes in orange, and microglia in light green.Water molecules (red circles) and their diffusion trajectories (red lines) are also shown as fitted ellipsoids (pink ellipsoids surrounded by red dotted lines).Changes in the FA and MD values associated with each pathology is represented by: up arrow, increase; no change, n.c.; decrease, down arrow.

Fig. 4 .
Fig. 4. Wallerian degeneration seen after a hypoxic-ischemic event.The subacute cerebral infarct area (red arrow) is depicted as a high-signal area on the mean diffusion-weighted image (DWI), with low mean diffusivity (MD) and low fractional anisotropy (FA).Wallerian degeneration (yellow arrow) in the left CST is hyperintense on DWI, with low MD and FA.

Fig. 5 .Fig. 6 .
Fig. 5. Graphical representation of changes over time in diffusivity measures seen in neonatal HIE.

Fig. 7 .
Fig. 7. Subacute ischemic changes seen after a hypoxic-ischemic event.The lesion (yellow arrow) is depicted as a high-signal area on mean diffusion-weighted images (DWI), showing low mean diffusivity (MD) and low fractional anisotropy (FA), probably representing a mixed lesion of cytotoxic edema and cell death.

Fig. 8 .
Fig. 8. Chronic changes seen after a hypoxic-ischemic event.The lesion (yellow arrow) is depicted as a slightly high-signal area on mean diffusion-weighted images (DWI), showing high mean diffusivity (MD) and low fractional anisotropy (FA), probably representing the gliosis that occurs after cell death.