Decreased oxygen saturation levels in neonates with transposition of great arteries: Impact on appearance of cerebral veins in susceptibility-weighted imaging

Purpose of this study was to investigate a potential correlation between the pattern of cerebral veins (CV) on susceptibility-weighted imaging (SWI) and blood oxygen saturation, as well as preoperative brain injury, in neonates with transposition of the great arteries (TGA). Eleven neonates with TGA underwent MRI preoperatively, including SWI, T1- and T2-weighted scans. Images were retrospectively evaluated and appearance of CV was graded from 0 (normal appearance) to 3 (severe prominent appearance). White matter injuries (WMI) and strokes were analysed. Results were correlated with preductal arterial oxygen saturation. As findings one subject showed a normal CV appearance (grade 0) whereas 10 showed pathological prominent CV (grades 1–3); median 2. Mean oxygen saturation ranged between 67.5% and 89.0% (median 81.0%). CV grade and mean oxygen saturation correlated significantly (p = 0.011). WMI were absent in 5 cases, mild in 4, and moderate in 2 cases. We conclude, that SWI has the potential to be used to estimate the current hypoxic burden on brain tissue in TGA newborns by assessing the prominence of the CV.

a phase mask, thereby emphasizing the magnetic properties of different susceptibility effects. SWI has improved diagnostic accuracy, e.g. through detecting and evaluating intracranial haemorrhage, haemorrhagic transformation of stroke or calcification [20][21][22] . In the cerebral veins (CV), decreased oxygen saturation correlates with an increase of de-oxygenated haemoglobin that leads to a prominent appearance of these veins 23 . This phenomenon is used for diagnostic purposes. Prominent CV were described in stroke patients in the perfusion-disturbed areas after thromboembolic occlusion [20][21][22][23][24] and in patients with hemiplegic migraine 25 due to hypoperfusion. The opposite effect, with diminished CV, was demonstrated in patients with hyperperfusion in status epilepticus and in hyperventilated patients 26,27 . A correlation between blood oxygen levels and prominence of CV has been also described 27,28 . Changes in blood oxygenation levels are well-known to also affect the signal intensity of cerebral veins in SWI 29 . Kitamura et al. introduced at first a scoring system of "prominence of vein" using SWI to asses neonatal hypoxic -ischemic injury. In this study a correlation of patients with "abnormal" "prominence of vein" and abnormal intermediate-term to long-term neurologic outcome was found 30 . Neonates with TGA suffer from low arterial oxygen saturation. It is assumed that the degree and duration of preoperative hypoxia is a risk factor for later neurocognitive deficits 8 . While white matter lesions in MRI scans reveal chronic brain injury, the prominence of CV in the SWI is a dynamic feature of the current shortage of oxygen supply.
Given these points, the aim of this study was to investigate whether a presumed general prominence of CV could be used diagnostically to estimate the current hypoxic burden on brain tissue. We hypothesized a correlation between the preoperative oxygen saturation in patients suffering from TGA and the prominence of CV in SWI, which reflects the deoxygenated share of haemoglobin. Further, a correlation between preoperative brain injury (amount of white matter lesions) and the grading of prominence of CV was analysed.

Materials and Methods
Patient data. The retrospective study was approved by the local Ethics Committee (Kantonale Ethikkommission Bern, Switzerland) and was carried out according to the Declaration of Helsinki 31 . An informed consent of the patients was not necessary due to the retrospective study design.
Subjects who met the inclusion criteria for this descriptive case study were term neonates with confirmed TGA who underwent routine preoperative MRI, and were hospitalized in our paediatric intensive care unit. Only neonates with TGA were included, because of the quite similar circulation anomaly in these patients, but variable blood oxygen saturation. During the MR examination all patients were sedated with either midazolam i.v. (total dose between 0.1 mg and 3.1 mg) and/or chloral hydrate p.o. (total dose between 60 mg and 300 mg). In individual cases rocuronium i.v. and/or ketamine i.v. was injected (up to 3.2 mg, and up to 3.5 mg, respectively). None of the children received propofol medication, because of the risk for propofol infusion syndrome. Mandatory sequences for this study were SWI, T1-and T2-weighted images. A further inclusion criterion was availability of values for arterial oxygen saturation (measured by means of both arterial blood gas analysis and pulse oximetry) and arterial carbon dioxide (CO 2 ) (measured by means of arterial blood gas analysis) that had been recorded on the same day that the patient underwent MRI. Concentrations of serum glucose in mmol/L, blood haemoglobin in g/L, blood pH, and heart rate in beats/min were noted. Further, for a higher objectivity three values of preductal peripheral oxygen saturation (SaO 2 ) during or close to MRI examination were determined: first the blood arterial oxygen saturation (blood gas analysis data closest to SWI acquisition time), second the single peripheral oxygen saturation measurement (pulse oximetry data closest to SWI acquisition time), and third the mean of peripheral oxygen saturation measurements, measured in between 2 h before and after SWI acquisition (pulse oximetry data). Patient data were excluded if image quality was poor, e.g. due to motion artifacts, if MRI data had not been acquired due to cardiovascular instability, or if the patient had a complex form of TGA, e.g. TGA with a single ventricle. Data acquisition. All imaging studies were performed using 1.5 T Siemens scanners (Magnetom Avanto and Magnetom Aera; Siemens Medical Solutions, Erlangen, Germany) with 12-channel head coils. The MRI protocol included axial SWI, T1 turbo spin echo (TSE) and T2 TSE. The following parameters were used. Data analysis. MRI sequences for all patients were reviewed independently by two board-certified neuroradiologists (RKV and DK for SWI, T1-weighted and T2-weighted images; RKV and CW for DWI). Other than knowing that these were TGA cases, they were blinded to the clinical course. MRI scans were assessed on our picture archiving and communication system (PACS) separately, in a standardized order: first SWI, followed by T1-weighted and T2-weighted images. When examining the SWI, both neuroradiologists classified the global prominence of CV according to 4 grades in the minimum intensity projection images (mIP): Grade 0: normal signal intensity, number, and diameter of cortical veins and deep veins (one or more veins up to 0.9 mm and 1.1 mm in diameter, respectively); no visible veins in the semioval centre. Grade I: slightly decreased signal intensity, increased number and diameter of cortical veins and deep veins (one or more veins up to 1.1 mm and 1.3 mm in diameter, respectively); virtually no visible veins in the semioval centre. Grade II: moderately decreased signal intensity, increased number and diameter of cortical veins and deep veins (one or more veins up to 1.4 mm and 1.7 mm in diameter, respectively); slightly to moderately visible veins in the semioval centre. Grade III: distinctly decreased signal intensity, increased number and diameter of hypointense cortical veins and deep veins (one or more veins up to 1.7 mm and 2.4 mm in diameter, respectively); distinctly visible veins in the semioval centre (see Fig. 1). This grading might vary depending on individual anatomical conditions, the strength of the magnetic field or the MR manufacturer, and therefore be imprecize. Hence, we included further typical characteristics of prominent CVA (increased number of cerebral veins, increased signal intensity of cerebral veins, increased diameter of cerebral veins, and visible veins in the semioval centre) for a more objective and differentiated grading. Further, for two reasons we did not exactly use the "prominence of vein" scoring system introduced by Kitamura et al. 30 , but modified it. For evaluation only the deep medullary veins were used in this system. Meanwhile many studies have revealed the usefulness of both cortical and deep veins for evaluation of vein prominence [22][23][24][25] . For a higher sophistication we included these veins in our scoring system. Secondly, in some recently published studies diminished cerebral veins occur in case of a higher cerebral blood flow or increased oxygen supply most likely resulting in higher oxygenated haemoglobin levels [26][27][28] . Since TGA patients suffer from low oxygen saturation levels, leading to increased deoxygenated haemoglobin levels, we have not applied Grade 1 of the Kitamura scoring (absent deep medullary veins).
For evaluation of white matter injuries (WMI) and cerebral strokes we used a score previously described by McQuillen et al. 13 . In T2-and T1-weighted images, WMI and white matter strokes were analysed and counted: WMI were defined as lesions measuring <2 mm of abnormal T1-hyperintensity or of low intensity on T1-weighted images in the absence of marked T2-hypointensity 13,16 . Severity of WMI was classified as absent (0 areas of T1 signal abnormality); mild (≤3 areas of T1 signal abnormality measuring <2 mm); moderate (>3 areas of T1 signal abnormality measuring <2 mm or areas of T1 signal abnormality measuring >2 mm but covering <5% of the cerebral hemisphere); or severe (T1 signal abnormality covering >5% of the cerebral hemisphere) (see Fig. 2A to C). Cerebral strokes were defined as focal regions of ≥3 mm with or without diffusion restriction on DWI 12 (see Fig. 2D to F). In a final session, a consensus reading was performed in cases of disagreement.
Statistical analysis. For statistical analysis the SPSS statistics software package was used (IBM, Armonk, N.Y., USA). A non-parametric, two-sided Kendall rank correlation coefficient test was applied to verify whether the two variables (prominence of CV and arterial blood oxygen levels) were statistically dependent. A p-value of <0.05 was considered significant. Median and total range for all parameters was determined (SWI grade, WMI, white matter strokes, arterial blood oxygen saturation, arterial blood pCO 2 , blood glucose, blood haemoglobin, blood pH, single monitored oxygen saturation measurement, and mean monitored oxygen saturation value).
For interrater reliability of findings in SWI, T1w and T2w sequences in MRI scans, a Cohens-Kappa test was performed and the interrater reliability was classified according to Landis and Koch 32 .
Data availability. The datasets generated during and/or analysed during the current study are not publicly available due to preclusion from dissemination following Swiss Federal Law Regulations, but are available from the corresponding author on reasonable request.

Results
Subjects and clinical data. Between March 2011 and July 2015, 11 patients (8 male and 3 female; mean age 5.7 days at time of MRI acquisition) fulfilled the inclusion criteria.
Blood CO 2 and blood haemoglobin were in the upper standard value range in 2 subjects. Blood glucose was normal in all patients.
Imaging findings. One subject showed a normal cerebral veins appearance (CVA) (grade 0) while 10 showed prominent CVA (4 patients with grade 1, 4 patients with grade 2, and 2 patients with grade 3); range between grade 0 and 3; median 2. Severity of WMI was graded as absent in 5 cases with no WMI, mild in 4, and moderate in 2 cases. Only one patient showed white matter strokes in the territory of the left middle cerebral artery with 3 lesions >2 mm. Furthermore, in one patient two acute ischaemic areas were found, in 3 patients 1 acute ischaemic area was found and 7 patients had no acute ischaemia (Table 1).
Patient No. 8 underwent follow-up MR imaging postoperatively, with an oxygen saturation of 95% and a normalized CV appearance Grade of 0 (see Fig. 3B). Three values of oxygen saturation were analysed (Fig. 4 for illustration): the blood arterial oxygen saturation (Fig. 4A, blood data closest to SWI acquisition time), single peripheral oxygen saturation measurement ( Fig. 4B; pulse oximetry data closest to SWI acquisition time), and mean of peripheral oxygen saturation measurements, measured ± 2 h of SWI acquisition (Fig. 4C; pulse oximetry data). A significant correlation was found for blood  Table 1. PFO, persisting foramen ovale; PDA, patent ductus arteriosus; ASD, atrial septum defect; VSD, ventricular septum defect; n.a., not available. Values of arterial O2-saturation in %, arterial pCO2 in mmHg, Glc in mmol/L, and Hb in g/L were taken from blood. Monitored SaO2 measurements (single measurement closest to SWI acquisition, and mean +/−2 hours of SWI acquisition) were pulse oximetry data. HR, heart rate in beats per minute.

Discussion
The main finding of our investigation is a significant correlation between the prominence of CVA and blood oxygen saturation levels. In our study only one newborn showed a normal appearance of the CV, whereas 10 revealed a pathological venous pattern with presumed higher deoxyhaemoglobin levels, significantly correlating with the arterial blood oxygen saturation (blood arterial oxygen saturation and monitored mean oxygen saturation). Only monitored single oxygen saturation measurements showed no significant correlation, because of an outlier value (patient No. 6 with 86% oxygen saturation and prominence of CV of grade 3). To the best of our knowledge this is the first report showing a significant correlation between blood arterial oxygen saturation levels and the extent of prominent CV in SWI of newborns suffering from TGA. However, in a recently published study by Öztoprak on 19 adult patients with pulmonary embolism, a significant correlation between presence of prominent veins on SWI and hypoxemia 33 was found, which is in line with our results. Further, our findings are in accordance with a recently published experimental study by Patzig et al. 34 . In their study 16 healthy adult volunteers underwent MRI under room air conditions, short-term hypoxia (7 minutes before and during the MRI scan), and long-term hypoxia (8.5 hours before and during the MRI scan). The 9 subjects who were finally included in the study showed significantly lower signal intensities of CV in the scans under conditions of hypoxia. However, our study is the first to investigate a patient population with chronically low arterial blood oxygen levels.
Blood CO 2 is another parameter with an influence on CVA. For example, in a recently published study, a correlation between hyperventilation (in which CO 2 is expelled leading to a vasoconstriction with lower cerebral blood flow) and higher signal intensity in CV was demonstrated, indicating a lower venous blood oxygenation level during hyperventilation 28 . Sedlacik et al. revealed a significant correlation between high etCO 2 and cerebral blood flow with weak venous contrast in SWI 35 in spontaneously breathing patients sedated with propofol. Since etCO 2 and blood bicarbonate levels show a significant correlation 36 , blood CO 2 might have influenced the CVA in our study. The difference between etCO 2 and PaCO 2 is stated to be 2-5 mm Hg with higher values for PaCO 2 37 .
Based on the findings of Sedlacik et al. etCO 2 should be kept as low as possible to obtain a good venous contrast in SWI. With a maximal etCO 2 value of 30-35 mmHg CBF in sedated children is not artificially elevated 38,39 . Although in our subjects, blood CO 2 was normal, with a median of 41 mm Hg it might have influenced the CVA, even by subtracting 2-5 mm Hg, since it is still higher than 30-35 mm Hg and might lead to an increase of CBF. However, in case of an elevated blood CO 2 and CBF a weak venous contrast would be expected. In our subjects a stronger venous contrast is observed, therefore we assume the influence of blood CO 2 is presumably low to negligible. Further blood parameters with a potential to lead to bias were in the normal range (median of blood glucose 5.2 mmol/L; median of blood haemoglobin 144 g/L; and median pH of blood, 7.37). These findings indicate that the changes in CVA are most likely caused by the blood oxygen saturation levels, correlating with other investigations 28,29,33,34 .
Many studies have demonstrated that newborns with complex CHD have abnormal brain development with brain injuries and delayed brain maturation [9][10][11] . Preoperatively, neonates with TGA exhibit brain injury which is Figure 3. SWI images of a patient (No. 8) as an example before (Fig. 3A) and after surgery (Fig. 3B). After surgery the CV appearance improved distinctly becoming normal. Before surgery oxygen saturation was 72% (CV appearance Grade II), after surgery oxygen saturation was 95% (CV appearance Grade 0). associated with hypoxia 15,17 . It can be assumed that if hypoxia improves in TGA patients, the neuroanatomical and developmental outcome improves too 40 . A technique currently used to reveal brain injury is MRI. Using diffusion-weighted T1w and T2w images, WMI and strokes can be detected. A correlation between WMI and neurological abnormalities has been described in neonates with severe CHD before surgery 16 . A study using magnetic resonance spectroscopy, which is an advanced neuroimaging technique, revealed an elevated lactate in neonates with TGA, indicating an abnormal brain metabolism 19 , even in subjects in whom MRI showed no evidence of preoperative brain injury. Diffusion tensor imaging studies revealed regions of reduced white matter 9,18 .
Our investigation showed no correlation of WMI, acute or chronic strokes, and SWI findings; e.g. patient No. 3 showed a clear discrepancy with 10 WMI lesions, but a mean oxygen saturation of 89% and a prominence of CV of grade 1. These results were expectable, since both the oxygen saturation measurement and the prominence of the CV provide a "snapshot" reflecting the current hypoxic burden, while WMI or strokes are unique events leading to chronic gliotic tissue damage. Nine of 11 subjects had no WMI, or only one lesion, but 10 of the 11 showed a pathological cerebral venous pattern.
The prominence of CVA in neonates with TGA in SWI is a new finding in this context, indirectly demonstrating an abnormal oxygen supply. Compared to chronic changes such as WMI, SWI has the advantage that it can demonstrate current acute changes in venous haemoglobin composition by CVA (see Fig. 3 as example). Therefore, SWI can potentially be used to estimate insufficient brain oxygen in neonates suffering from TGA, which might be important e.g. in postoperative control MRI to assess whether brain oxygen supply has improved. A relatively new method, for assessing cerebral oxygen saturation in neonates, is near-infrared spectroscopy (NIRS) 41 . Both techniques evaluate the "venous weighted" cerebral haemoglobin values. NIRS has some clear advantages. It evaluates the cerebral oxygen saturation, is non-invasive, measures cerebral oxygen saturation continuously, and is as such a valuable bedside tool. However, NIRS has some limitations including a limited penetration depth of the NIR light, i.e. reaching the cerebral cortex only, the instability of NIRS values, and the lack of a direct reference against which to correlate the values 42 . This might lead to a false higher cerebral oxygen saturation values in individual cases. Further, the derived NIRS-signal contains extracerebral "contamination" from bone and skin tissue. SWI has the ability to detect prominent cerebral veins in the whole brain in a standardized manner and therefore provides additional clinically meaningful information, despite the more work-intensive data acquisition. Since a brain MRI in children with congenital heart disease is carried out anyway, SWI in this context can be acquired quite easily.
Our study does however have limitations. Cerebral metabolic rate of oxygen and cerebral blood flow were not measured during MRI acquisition. Although proprofol was not applied to our patients, we cannot fully rule out an influence on cerebral metabolism and cerebral blood flow, caused by an anesthesic used in our patients, as described in previous studies for midazolam, chloral hydrate or ketamine [43][44][45] . Midazolam can slightly decrease cerebral blood flow 43 , which might contribute to the prominence of CV, because of higher deoxygenated hemoglobin levels. In contrast, ketamine might lead to a mild increase of cerebral blood flow with no effect on the metabolic rate 45 , which might contribute to lesser prominent CVs. For chloral hydrate both, a decrease of cerebral blood flow (resulting in higher deoxygenated hemoglobin levels) and a decrease of cerebral glucose utilization (resulting in higher oxygenated hemoglobin levels) was described 44 . Hence, an effect on the prominence of CV on SW images is more likely neutralized.  Figure 4B shows no significant correlation between SWI grade and SaO2 in % of single peripheral pulse oximetry data, because of an outlier: Pt. No. 6 with SWI grade 3 and 86% SaO2 (p = 0.147; Kendall's tau = −0.401; regression line y = 87.29-5.43 * x). Figure 4B reveals a significant correlation between SWI grade and SaO2 in % of mean peripheral pulse oximetry data (p = 0.011; Kendall's tau = −0.680; regression line 89.03-5.98 * x).
Further, using a retrospective study design we analysed a standard SWI sequence and graded the prominence of CV visually. Although the outcome was in line with our hypothesis, a more objective quantification of venous prominence would be helpful. An example of a relatively new approach is quantitative susceptibility mapping (QSM), using SWI for measuring oxygen saturation; it was introduced by Haacke et al. 46 and has the potential to provide quantitative estimations of oxygen saturation in CV. This technique uses phase images and has been applied to CV in stroke patients for estimation of oxygen saturation in a recent study 47 . Both SWI and QSM are gaining increasing acceptance in clinical practice 48 . In future studies, acquisition of SWI with phase images in neonates with TGA would help to quantify the extent of oxygen shortage in CV.
Another limitation of the present study is the small number of patients. A larger cohort is needed to confirm our results.
In conclusion, the appearance of prominent cerebral veins in SWI is a pathological finding in 10 of 11 newborns with TGA in our study. It indicates an increase of deoxygenated hemoglobin, and thus lack of oxygenated hemoglobin. Our study demonstrated a significant correlation between the appearance of CV in SWI and the arterial blood oxygen levels in neonates with TGA. SWI has the potential to be used to estimate the current hypoxic burden on brain tissue by the effect on the appearance of CV. As a dynamic sequence SWI reflects the oxygen shortage in a "snapshot", while other conventional sequences e.g. T1w or T2w images show chronic or subacute ischaemic changes. Therefore SWI might be a helpful additional sequence to the MRI protocol for TGA patients, to clarify the possibility and the extent of cerebral hypoxia (e.g before and after surgery), and rule out other intracranial pathologies, like hemorrhagic transformation of ischemic regions. Further studies with a larger cohort and SWI phase images would be beneficial to test our findings objectively by quantifying the cerebral venous oxygen saturation.