How accurate are prenatal tractography results? A postnatal in vivo follow-up study using diffusion tensor imaging

This study is a retrospective observational study conducted between April 2006 and July 2015. The inclusion criteria for this study were the presence of both prenatal and postnatal MRIs, performed at the Department of Biomedical Imaging and Image-guided Therapy of the Medical University of Vienna, Austria, and no interventions to the brain in the interval of the two MRI examinations. We identified 75 subjects with both fetal and postnatal MRIs. We checked each examination for DTI sequence acquisitions. Fourteen fetal MRs and 32 postnatal MRs resulted in 13 subjects with both fetal and postnatal MRs with DTI sequences. After assessing the 13 prenatal DTI datasets, we excluded one case because of poor signal to noise. Thus we included 12 cases with paired fetal and postnatal MRI with DTI.

The 12 subjects comprised 7 girls and 5 boys. Gestational age was determined by biometry during the first sonographic examination and is presented as postconceptional age. The mean gestational weeks of age of the fetuses was 30 and 1/7 weeks (range: 23 and 3/7 weeks to 35 and 2/7 weeks). The mean postnatal age was 5 months (range: 1 day to 2 years). The mean interval time between the fetal and postnatal DTI exams was 262 days (range: 12–1,081 days). Eight cases had primary central nervous system (CNS) pathologies, three had secondary CNS pathologies and one had a non-neurological congenital diagnosis (Table 1). The categorization of primary versus secondary CNS pathologies was based on whether the pathology was derived from a neurologic origin or was a result of a more systemic pathology. For instance, pathologies defined as primary CNS pathology included lissencephaly or septo-optic dysplasia. A secondary CNS pathology included middle cerebellar artery vascular infarct in the context of congenital heart disease (tetralogy of Fallot). Only one case was thought to be in the “other” category, with no known obvious CNS pathology; that case was imaged for the clinical condition of gastroschisis.

Data were obtained from pregnant women who were informed about the MR examination and possible risks. The ethics committee of the Medical University of Vienna approved the protocol and waived written informed consent because of the study’s retrospective design. We anonymized all image data by case number before further analysis.

Prenatal imaging was performed on a 1.5-T MR (Gyroscan Intera; Philips Healthcare, Best, the Netherlands) and a five-element phased-array channel cardiac coil. In addition to the standard structural fetal MR sequences (multiplanar T2-W single shot turbo spin echo [TSE], echo time [TE] 140 ms, repetition time [TR] shortest, field of view [FOV] 230–260 mm, flip angle 90°, matrix 256x153, voxel size 0.78/1.18 /4.00 mm; multiplanar steady-state free precession [SSFP], FOV 260 mm, matrix 192x219, voxel size 1.35/1.19/3 mm, flip angle 80°), a fetal DTI sequence was acquired. The axial single-shot fat-suppressed echoplanar DTI sequence included 16 noncollinear diffusion gradient-encoding directions with b values of 0 s/mm² and 700 s/mm², TE 90 ms, TR shortest, flip angle 90°, matrix 112x105, slice thickness 3 mm and acquisition time of 1 min 16 s. The voxel size was 2.8–4.3 mm³. No sedation or intravenous contrast agents were administered. Postnatal imaging was performed on either the same 1.5-T scanner as the prenatal examination (Intera; Philips Healthcare, Best, the Netherlands; TR/TE 4,371/90 ms, b values of 0 s/mm² and 700 s/mm², 16 gradient encoding directions, slice thickness 4 mm, matrix 128x128, acquisition time 4–5 min) or on a Magnetom Aera 1.5 T (Siemens Healthcare, Erlangen, Germany; TR/TE 9,600/120 ms, b values of 0 s/mm² and 700 s/mm², 16 gradient encoding directions, matrix 1,072x1,072, slice thickness 2–3 mm, acquisition time 5–6 min). The voxel size was 0.83 mm³. An 8-channel head coil was used for postnatal imaging. All postnatal examinations were performed under sedation and in the presence of an anesthesiologist with dedicated experience in pediatric neuro MRI.

A multiple regions-of-interest (ROI) approach was used to reconstruct the corticospinal tracts and corpus callosum using anatomical knowledge of the tract trajectories as previously described in published protocols [10, 11]. The corticospinal tracts and corpus callosum were segmented using the Philips Intellispace Portal system by two radiologists (J.W.S., with 1 year of experience, and G.K., with 3 years of experience) who were blinded to either prenatal or postnatal tractography to ensure no bias in identifying tract morphology at the different ages.

We performed qualitative analyses including evaluating the presence of each tract at both stages. We evaluated presence as a dichotomous variable — present or not present — by assessing for the presence of fibers accounting for the degree of signal to noise. We also evaluated the corpus callosum by its anatomical parts: rostrum, genu, body and splenium. The rostrum was identified as a beak-shaped segment curving posteriorly from the genu. The genu was identified as the curved anterior portion of the corpus callosum projecting anteriorly to approximately the region of the line connecting the mammillary body, anterior commissure and corpus callosum [12]. The body of the corpus callosum was identified as a horizontal structure in the expected location of the corpus callosum on midline sagittal images. Finally, the splenium was identified as a caudally oriented or bulbous posterior portion [13].

We categorized the morphology and integrity of the corticospinal tracts into the following four categories: (1) not visualized, (2) discontinuous, (3) intact and (4) artifact. We performed this categorization visually by assessing the presence and morphology of the fibers accounting for the degree of signal to noise.

Next we evaluated the utility of fetal DTI as a diagnostic test for assessing corpus callosum and corticospinal tract pathways by tractography. We defined true positives as cases where pathways were identified in both fetal and postnatal DTI by tractography. True negatives were cases where pathways were not identified on both the fetal and postnatal DTI by tractography. False positives were cases where pathways were visualized by fetal tractography but were not present on postnatal tractography. Finally, false negatives were cases where pathways were not visualized on fetal tractography but were present on postnatal tractography. Accuracy was calculated as the total number of true positives and true negatives divided by the sum of true positives, true negatives, false positives and false negatives. We calculated positive predictive values (PPV) as the number of true positives divided by the total number of true positives and false positives. Negative predictive values (NPV) were the number of false positives divided by the total number of true positives and false positives.

We also compared quantitative outcomes including fractional anisotropy (FA) and apparent diffusion coefficient (ADC). We measured the ventricular sizes to evaluate for the degree of ventriculomegaly. The diameter of the atria of the lateral ventricles was measured on coronal T2-weighted images [14]. Ventriculomegaly was defined as an atrial diameter exceeding 10 mm.

Anatomical segments (rostrum, genu, body and splenium) of the corpus callosum were evaluated as being present or not present by tractography (Figs. 1 and 2, Table 2). Among the callosal segments, the body of the corpus callosum was visualized the least frequently by tractography; it was only visualized 17% (n=2 of 12) of the time prenatally and 25% (n=3 of 12) of the time postnatally. The rostrum was visualized 42% (n=5 of 12) of the time prenatally and 75% (n=9 of 12) of the time postnatally. The genu was visualized 67% (n=8 of 12) of the time prenatally and 92% (n=11 of 12) postnatally. Finally, the splenium was visualized 58% (n=7 of 12) of the time prenatally and 75% (n=9 of 12) postnatally.

We also assessed the presence of the corpus callosum by segments on midline sagittal T2-weighted images of the brain on fetal and postnatal MR exams and compared it to the presence or absence of the segmented tractography by DTI. There was a moderate to strong correlation between the identifiable segments by MR and the tractography. Table 3 shows the results of the MR evaluation with respect to the tractography findings; for instance, when the rostrum was visible on the MR exam, it was also evaluated by tractography and was categorized as a “matched” result if seen on both anatomical MR and DTI tractography exams. On the fetal MR exams, the tractography results matched the anatomical MR exam 75% (n=9 of 12) of the time for the rostrum, 83% (n=10 of 12) of the time for the genu, 58% (n=7 of 12) of the time for the body and 75% (n=9 of 12) of the time for the splenium. On the postnatal MR exams, the tractography results matched the MR imaging findings 83% (n=10 of 12) of the time for the rostrum, 100% (n=12 of 12) of the time for the genu, 50% (n=6 of 12) of the time for the body, and 83% (n=10 of 12) of the time for the splenium.

Accuracy values for the genu and body were 75% (n=9 of 12), for the rostrum 67% (n=8 of 12), and for the splenium 67% (n=8 of 12; Table 2). These results suggest that the presence or absence of segments of the corpus callosum by tractography on fetal DTI is moderately accurate for the presence of these fiber pathways by postnatal DTI imaging, even in the setting of underlying neuropathology. There were a 100% PPV for the rostrum and genu, 50% PPV for the body and an 87.5% PPV for the splenium (Table 4). Negative predictive values ranged from 25% to 80%, as shown in Table 4. These findings suggest a very high degree of PPV for prenatal tractography of the corpus callosum.

Figure 3 illustrates a case with a large midline parieto-occipital cyst exerting mass effect on the corpus callosum and thereby displacing the fibers inferiorly and deforming it. Fetal tractography was deemed accurate for all four segments of the postnatal segmentation. For example, in the fetal and postnatal segmentations, the rostrum, genu, body and splenium were all visible by tractography. Consequently, this favored excluding callosal dysgenesis or partial callosal agenesis as a descriptive diagnosis prenatally and impacted counseling.

Figure 4 illustrates a case of periventricular leukomalacia as manifested by subtle periventricular T2 prolongation. At the fetal tractography, the rostrum and body were not fully present. Conversely, in the 2-year-old postnatal DTI all segments of the corpus callosum were present by tractography. In this example, only the genu and splenium were accurate.

Figure 5 illustrates a case of a right middle cerebral artery infarct that resulted in cystic encephalomalacia of the right middle cerebral artery vascular territory. Figure 5 shows a paucity of fibers in the body and splenium of the corpus callosum by fetal tractography. In the 4-month-old postnatal brain, the body of the corpus callosum is also not visualized by tractography. Thus the absence of the body of the corpus callosum by fetal tractography was accurate because it was not seen on postnatal tractography either. However the absence of the splenium in the fetal tractography was not accurate because it was present at postnatal tractography.

We performed a subgroup analysis comparing whether a primary neurologic abnormality (e.g., case 3, lissencephaly) versus secondary (e.g., case 10, cortical infarcts related to intrauterine demise of fetal twin) or unrelated (e.g., case 1, gastroschisis) pathology resulted in less accurate results. Among cases with primary neurologic pathology 50% (n=4 of 8) revealed three or more accurate results for the four segments of the corpus callosum. The mean PPV was 100% accounting for all four segments. The mean NPV was 50%. In cases with secondary or unrelated neurologic pathology, 100% (n=4 of 4) revealed three or more accurate results between fetal and postnatal segmentations of the corpus callosum. The mean PPV was 92%. No mean NPV was calculated because no true-negative cases for the genu and splenium were present.

We also performed a second subgroup analysis, by fetal age (younger or older than 28 gestational weeks). Four cases comprised fetuses younger than 28 gestational weeks, and eight cases comprised fetuses older than 28 gestational weeks. Among the fetuses younger than 28 gestational weeks, 75% (n=3 of 4) of the cases revealed three or more accurate results for the four segments of the corpus callosum. The mean PPV was 70% accounting for all four segments. A mean NPV was not calculated because no true-negative cases for the rostrum, genu or splenium were present. In fetuses older than 28 gestational weeks, 63% (n=5 of 8) of the cases showed three or more accurate results among the four segments of the corpus callosum. The mean PPV was 95% accounting for all four segments. A mean NPV was 58%.

Apparent diffusion coefficient and FA were graphed by paired cases. Visually, the cases demonstrated similar fetal and postnatal outcomes for ADC and FA (Fig. 6).

The right and left corticospinal tracts were segmented separately. The PPV was 100% for the right corticospinal tract and 88.9% for the left corticospinal tract. The NPV was 50% for the right and 33% for the left corticospinal tracts. Accuracy values ranged up to 92% (n=11 of 12) for the right and 75% (n=9 of 12) for the left corticospinal tracts (Tables 4 and 5).

Figure 7 illustrates segmented right (yellow fiber tracts) and left (green fiber tracts) corticospinal tracts in a case with a midline parieto-occipital cyst. Both corticospinal tracts were visualized on fetal and postnatal tractography and appeared to be intact although displaced posteriorly by the large cyst. The blue fiber tracts represented the callosal fibers. The prenatal segmentation of both the right and left corticospinal tracts was accurate of the postnatal segmentations.

Figure 8 illustrates the corpus callosum and right and left corticospinal tracts in the fetal and 2-year-old images in a child with periventricular leukomalacia. The right corticospinal tract in the fetal segmentation appeared intact. However the left corticospinal tract appeared discontinuous. By contrast, the corticospinal tracts in the 2-year-old postnatal brain appeared intact bilaterally (Fig. 8). Noted was a slightly more sparse appearance of the right compared to the left corticospinal tract, which was thought to represent Wallerian degeneration as manifested by the severe appearance of white matter injury in the right cerebral hemisphere (Fig. 8).

Figure 9 illustrates a case of a large right middle cerebral artery vascular territory infarct with ex vacuo dilatation of the right lateral ventricle. Fetal DTI tractography revealed no identifiable right corticospinal tract fibers, only left corticospinal tract fibers. In the postnatal DTI, a few residual fiber tracts composing the right corticospinal tract were noted, possibly reflecting atrophic or residual fiber tracts given their sparse appearance (Fig. 9). This appearance was markedly different from the left corticospinal tract in both the fetal and postnatal DTI segmentations. The prenatal segmentation of both the right and left corticospinal tracts was accurate with regard to the postnatal segmentations.

We analyzed whether a primary brain pathology versus secondary or unrelated (e.g., case 1, gastroschisis) brain abnormality decreased accuracy. Among cases with primary neurologic pathology (n=8), accuracy values were 88% (n=7 of 8) for the right corticospinal tract and 75% (n=6 of 8) for the left corticospinal tract. Positive predictive values were 100% for the right corticospinal tract and 83% for the left corticospinal tract. The NPV was 50% for the left corticospinal tract. No true-negative case emerged for the right corticospinal tract and an NPV could not be calculated. Among cases with secondary or unrelated pathology, 100% (n=4 of 4) of the cases revealed accurate results between fetal and postnatal segmentations for the right and 75% (n=3 of 4) for the left corticospinal tracts. Positive predictive values were 75% for the right and 100% for the left corticospinal tracts. Negative predictive value was 100% for the right corticospinal tract. No true-negative case among the four cases with secondary or unrelated pathologies was present for the left corticospinal tract and an NPV could not be calculated.

We also performed a subgroup analysis by fetal age (younger or older than 28 gestational weeks) for the corticospinal tracts. Among the fetuses younger than 28 gestational weeks, 100% (n=4 of 4) showed accurate results for the right and 75% (n=3 of 4) for the left corticospinal tracts. The PPVs were 100% for both the right and left corticospinal tracts. No true-negative case emerged and an NPV could not be calculated for either the right or left corticospinal tract. Among fetuses older than 28 gestational weeks, 88% (n=7 of 8) showed accurate results for the right and 75% (n=6 of 8) for the left corticospinal tracts. The PPVs were 100% for the right and 83% for the left corticospinal tracts. The NPVs were 50% for both the right and left corticospinal tracts.

Apparent diffusion coefficient and fractional anisotropy (FA) were visually graphed by paired cases. For the left corticospinal tract, the paired prenatal and postnatal cases revealed similar ADC and FA outcomes for the right and left corticospinal tracts (Figs. 10 and 11).

The corpus callosum, the largest commissural fiber tract, connects the two cerebral hemispheres. Studies have shown that dysgenesis or agenesis of the corpus callosum alters the structural connectome of the developing brain [4, 17]. Thus early detection of abnormal development of the corpus callosum is important to recognize during pregnancy because it could herald other underlying developmental abnormalities. Prenatal tractography results of the corpus callosum were accurate in more than two-thirds of the examined paired cases (Table 4), with accuracy values ranging between 67% and 75% and PPVs ranging between 50% and 100% for the four segments of the corpus callosum. These results suggest that if the corticospinal tract fiber pathways are visualized on prenatal DTI exams with a very high degree of confidence, the corticospinal tracts are likely to also be visualized on postnatal exams.

Interestingly we found poor visualization of the body of the corpus callosum, not only in prenatal but also postnatal tractography exams. This structure showed the most significant inconsistencies with structural imaging, where it could be visualized. The body of the corpus callosum is the thinnest segment of the corpus callosum, with a thickness in fetal brains at 24 gestational weeks of approximately 1.9 mm (range: 1.3–2.5 mm for the 5th–95th percentiles) as assessed by fetal ultrasound [18]. Also postnatally (0–2 months), the thickness of the mid-body of the corpus callosum has been described to be approximately 2.3 mm [19]. One study evaluated 152 neonates by ultrasound from ages 28 gestational weeks to 41 gestational weeks and reported no significant change in the thickness of the body of the corpus callosum, which measured approximately 2.7±0.4 mm [20]. In our study, with slice thicknesses of 3–4 mm for fetal DTI and 2–4 mm on postnatal DTI on 1.5-T MRI, diffusion information might have been lost because of partial volume effects in that specific region. Moreover, in four of the cases the corpus callosum was further thinned by hydrocephalus or ventriculomegaly, limiting the success of DTI in visualizing this segment. Notably, the body of the corpus callosum was not visualized in all four of these cases, yielding an accuracy of 100% in paired fetal and postnatal DTI tractography in cases with this structural morphology. These values contrasted with lower accuracy (63%) in cases without ventriculomegaly. It is, however, also important to note that in one case the body of the corpus callosum appeared to be absent on postnatal imaging, indicating that the absence of fiber pathways by tractography of the body of the corpus callosum in both the fetal and postnatal imaging might in some cases be true developmental absence of this structure (Fig. 2).

Interestingly, in two cases segments of the corpus callosum that were visualized by fetal tractography were not visualized in the postnatal exams (cases 8 and 11). Case 8 had a left middle cerebral artery infarct in the setting of tetralogy of Fallot; the body of the corpus callosum was visualized on fetal DTI but not on the postnatal DTI. Case 11 presented with septo-optic dysplasia and aqueductal stenosis; the splenium was visualized on fetal tractography but not on postnatal tractography. Explanations in these instances might be related to evolving changes of the original pathology (e.g., infarct or hydrocephalus) during development, sequelae of which might manifest more distinctly with maturity and development. Thus for certain types of injuries that continue to evolve during development, it is important to recognize that the fiber tracts might be restructuring as well.

The corticospinal tracts revealed accuracies of 75–92% between paired fetal and postnatal tractography results. Similarly, the PPVs were 100% for the right and 88.9% for the left corticospinal tracts. These results suggest that if fiber pathways of the corticospinal tracts are visualized on prenatal DTI exams with a very high degree of confidence, the corticospinal tracts are also likely to be visualized on postnatal exams. The higher degree of accuracy of the corticospinal tracts compared to the corpus callosum could be related to a more advanced and mature stage of the fibers. Corticospinal tracts begin myelination at about 34 gestational weeks, whereas studies have suggested that callosal fibers do not myelinate until late third trimester [21‐23].

Subgroup analyses assessing the type of injury (e.g., primary versus secondary/other, as defined in the methods section) did not reveal major differences in the accuracy of prenatal and postnatal tractography. Similarly, age (older or younger than 28 gestational weeks) did not appear to influence the results. However a larger sample size might be needed to better evaluate these potential confounders.

There are several limitations to this study. First, a few examinations had poor signal-to-noise ratio (SNR), which might have been minimized by using a 3-T MRI (Figs. 12 and 13). Although fetal MRI exams are presently routinely performed on a 1.5-T MRI, 3-T MRI can increase SNR, decrease acquisition time and increase spatial resolution [24, 25]. Motion degradation is another commonly encountered artifact in fetal MR imaging. The mothers are not sedated for the MR examination, and there are few other methods of minimizing fetal motion apart from altering the parameters to obtain faster acquisitions. Naturally, there are limitations to shortening acquisition times from the trade-off of SNR or spatial resolution.

Because of the retrospective nature of the study and the clinical indications of the respective exams, we included postnatal follow-up data from two 1.5-T scanners using standard neonatal DTI scanning parameters and acquisition times of about 4–6 min. However in comparing the data from both scanners there was no visible difference in the tractography results of the corpus callosum and corticospinal tracts.

A third limitation is the small sample size. Only 12 subjects met the inclusion criteria for the study with both fetal and follow-up postnatal MR imaging with DTI sequences. The high dropout rates can be explained by the fact that only cases with prenatally suspected brain abnormalities underwent postnatal MR imaging. At our institution early postnatal MRI is not routinely performed in cases without postnatal neurological deficits and with minor or normal CNS findings. Moreover, in cases with a poor neurodevelopmental prognosis, parents frequently opt for termination of pregnancy.

Another limitation is the variation in length of the interval time between fetal and postnatal imaging among the cases. The interval time ranged from 12 to 1,081 days. It is conceivable that cases with longer interval periods between the fetal and postnatal imaging and with earlier fetal ages might show in different results. The earliest fetal age was 23 gestational weeks, at which point the studied white matter tracts have developed. Nevertheless changes in the configuration of the white matter pathway and ongoing axonal growth certainly have influenced the appearance of the examined tracts. In order to detect and quantify these developmental changes, the inclusion of more cases and a study design using an atlas-based approach is required.

Last, given the 9-year span of the data collection, it is conceivable that the imaging parameters were optimized over time. Hence, a DTI examination from 2006 might have been of different quality from an examination performed in 2015. However, different parameters and coils were used for the fetal and postnatal DTI examinations, so we think this aspect of the retrospective study is noncontributory given the paired comparisons were already imaged differently. Moreover, in an attempt to objectively segment the fiber tracts two blinded investigators performed the segmentation on the corpus callosum and corticospinal tracts for each pair.