In this study of 121 extremely premature infants around term-equivalent age, no SDH or other extra-axial haemorrhage or blood products were identified either supratentorially or infratentorially on T1-weighted, T2-weighted, or SWI sequences. Nearly half of the infants had markers of a previous intracerebral haemorrhage related to their prematurity, including isolated germinal matrix-intraventricular haemorrhages, germinal matrix-intraventricular haemorrhages with sequelae of porencephalic cysts, and cerebellar haemorrhages. Inter- and intra-observer agreement for identification of haemorrhage was good and very good, respectively. The mean sinocortical width was 3.4 mm (3.5 mm on the right side and 3.3 mm on the left side), whereas the mean interhemispheric distance was 3.1 mm, both measured on axial T2-weighted MRI. Seventy-four infants (61.1%) had a sinocortical width > 3 mm on one or both sides, while only two infants (1.6%) had an interhemispheric distance > 6 mm.

Very few studies have reported on the prevalence of SDH in infants born prematurely [14, 39]. In a study by Keenan and colleagues including 152 cases of serious or fatal inflicted traumatic head injury, the authors found a mildly increased odds ratio for inflicted brain injury and SDH in children born prematurely [39]. Hitherto, only one study by Högberg and colleagues, being a population-based register-study of 306 infants, has addressed infants born before 28 gestational weeks [14]. The study included all infants born between 1997 and 2014 diagnosed with SDH by using the ICD-10 diagnosis of either S06.5 (traumatic SDH) or I62.0 (nontraumatic SDH), with no mention of location, appearances, or method of identification. This clearly reduced the study’s potential to recognise birth-related posterior fossa SDHs. The overall incidence of SDH was 16.5 per 100,000 infants, and the median age was 2.5 months. For infants aged 7–365 days, acute nontraumatic SDH was associated with multiple birth, preterm birth, and small-for-gestational age. Infants born prior to gestational week 32 had odds ratios of 1.96 for traumatic SDH, 3.21 for nontraumatic SDH, and 4.17 for SDH and an abuse diagnosis. The authors conclude that the increased odds for acute nontraumatic SDH in preterm births or small-for-gestational age indicate a perinatal vulnerability for SDH beyond the first week of life. Our findings do not support this hypothesis, as we did not find any evidence of SDH around term-equivalent age in our group of 121 infants born before 28 gestational weeks. Högberg and colleagues acknowledged several limitations to their study though, including a lack of validation of the SDH-diagnoses. They state that they used “observation for suspected abuse”, “battered baby syndrome”, or “maltreatment syndrome” as definition of an abuse diagnosis, but they did not specify whether (or which) specific ICD-10 codes were used. Neither did they address whether it is common practice to set an ICD-10 abuse-diagnosis in Sweden in cases of suspected or confirmed abuse. We consider these to be additional limitations, as both the use and predictiveness of ICD-10 codes for child abuse have been shown to vary considerably in clinical practice [40]. Given that our findings do not support the conclusions made by Högberg and colleagues and considering the limitations of their study, their conclusions must be read with caution.

In our study, given the infants’ mean chronological age of 14.7 weeks at the time of the MRI examination, the appearances of a potential SDH would have differed according to its developmental stage and potential rebleeds and to the type of MRI contrast used (i.e. T1- or T2-weighted). Our protocol was designed to detect both acute, subacute, and chronic SDHs, as well as markers of a previous bleed (hemosiderin), using T1-weighted, T2-weighted, and SWI. However, and opposite to parenchymal brain haemorrhages, hemosiderin deposits may or may not be present after a SDH [41, 42]. It has been suggested that the absence of a blood–brain barrier in the subdural space allows for dilution and clearance of hemosiderin [41, 42], hampering the sensitivity of hemosiderin deposits as a marker of SDHs. In term infants, birth-related SDHs are relatively prevalent, with most resolving completely within 1–3 months [12, 13], and they rarely go on to form chronic SDHs [12, 13, 43]. Little has been published on infants born extremely premature regarding this. Assuming that the prevalence and development of SDHs in infants born extremely premature may resemble that of term-born infants, it is possible that some of the infants could have had small previous SDHs that were fully reabsorbed at the time of the MRI scan around term-equivalent age, with no remaining hemosiderin deposits.

Compared to infants born at term [20,21,22], a larger percentage of infants born prematurely have prominent subarachnoid spaces at term-equivalent age [23,24,25]. Infants born prematurely are at risk of various encephalopathies including periventricular leukomalacia, germinal matrix- and intraventricular haemorrhage, parenchymal venous haemorrhagic infarction, and post-haemorrhagic hydrocephalus [23, 26], which may result in brain atrophy and prominent subarachnoid spaces [24, 27, 28]. It has also been suggested that premature birth in itself might give reduced brain growth, resulting in prominent subarachnoid spaces [29]. In the setting of BESS, it has been suggested that the enlarged subarachnoid spaces predispose to the development of spontaneous SDHs or SDHs after minor head trauma [15,16,17,18,19] due to increased stretching of the bridging veins, making them more vulnerable to tearing and subsequent bleeding into the subdural space [44, 45]. This hypothesis has since been challenged by the results of Raul and colleagues, suggesting that the enlargement of the subarachnoid space may actually dampen the displacement of the brain relative to the skull during motion, thereby reducing the stretch on bridging veins during brain-motion within the skull [46]. The pathophysiology of prominent subarachnoid spaces in ex-prematures might be different from that of BESS. However, some similarities exist, such as a longer route for the traversing vessels with potential stretching, which has been demonstrated by a mathematical model [30]. Nevertheless, the association between a greater depth of the subarachnoid space and an increased risk of SDHs is controversial [15, 31].

Studies that provide population-based references of subarachnoid space measurements for both infants born at term and infants born prematurely typically rely on cranial coronal ultrasound measurements [20,21,22, 24, 29, 47, 48] rather than on axial MRI, as performed in our study. However, linear measurements of the subarachnoid space obtained from cranial ultrasound in the coronal view have shown a good correlation with measurements obtained from coronal MRI [25, 49]. Further, a recent study, including 63 infants from 4 days to 24 months of age, found that the width of the subarachnoid spaces was slightly larger when measured on coronal T2-weighted images as compared to axial T2-weighted images [50]. Although based on only a small subset of our cohort (n = 10) who underwent coronal FLAIR imaging, the measurements of the subarachnoid spaces obtained from coronal FLAIR sequences were larger than those obtained from axial T2-weighted images. Taken this into consideration, our reported values are rather conservative as the subarachnoid spaces probably would have been slightly wider if measured coronally.

In term newborns, the mean sinocortical width measured on cranial ultrasound has previously been reported from 1.2–2.0 mm [20,21,22], which is significantly lower than our mean sinocortical width of 3.4 mm. A sinocortical width of 3.0 mm and an interhemispheric distance of 6.0 mm have been proposed as upper limits to distinguish normal from pathologically dilated subarachnoid spaces in infants, as measured on coronal ultrasound [21]. In our study, 61.1% had a sinocortical width > 3.0 mm on one or both sides, and only 1.6% had an interhemispheric distance > 6.0 mm, as measured on axial MRI images. According to the aforementioned paper [50], our figures are most likely underestimating the coronal width; thus, a large percentage of the ex-premature infants in our study have prominent subarachnoid spaces as compared to infants born at term. Since increased width of the subarachnoid space represents an important marker of underlying pathology, we suggest that reference standards based on different imaging methods and planes might prove useful.

There is limited published data on the width of subarachnoid spaces in infants born prematurely, measured at term-equivalent age. However, Vo Van and colleagues evaluated the subarachnoid space dimensions (craniocortical width and interhemispheric distance) using both cerebral MRI and ultrasound at term-equivalent age in infants born prematurely before 32 weeks of gestation [25]. They found a mean craniocortical width measured on coronal MRI of 5.0 mm on the right side and 5.1 mm on the left side, and mean craniocortical width measured on ultrasound of 3.3 mm on the right side and 3.0 mm on the left side, respectively. The interhemispheric distance was reported at 4.8 mm measured on MRI and 4.1 mm measured on ultrasound. As the craniocortical width is generally considered to be slightly wider than the sinocortical width [21] and we measured sinocortical rather than craniocortical width, their measurements are not directly comparable to ours. However, their interhemispheric distance measurements are slightly larger than those observed in our study. Regarding BESS, there is currently no consensus on imaging criteria and no established cut-off values, but age-dependent sinocortical-, craniocortical-, and interhemispheric widths above the 95th percentiles are considered abnormal [51].

The strengths of this study are the relatively high number of examined infants, its population-based design, state-of-the-art MR imaging, the meticulous assessment of the images by highly experienced observers, and the high agreement between observers for diagnosing haemorrhage/blood products. We do acknowledge limitations to our study, however. We did not obtain clinically measured head circumferences at the time of MRI conducted around term-equivalent age, which would have been relevant for several reasons. These measurements could have provided valuable clinical-radiological correlations within this cohort and allowed for a clearer demonstration of differences between ex-premature infants with prominent subarachnoid spaces and those with BESS in future studies. This is particularly important because a significant percentage of the infants in our study had prominent subarachnoid spaces as measured on MRI. Due to the “feed-and-wrap” technique used, some of the MRI examinations were aborted because of an uneasy or moving child. Thus, the SWI sequence was unsuccessful in 23 infants, which might have led to missed small hemosiderin deposits based on T2-weighted images alone. A coronal FLAIR sequence was added in 10 patients only, and no other coronal sequences or reconstructions were performed, with the potential of missing subtle SDHs. In addition, 13.0% of the examinations were considered of suboptimal quality mostly due to motion artefacts. Nevertheless, no appreciable SDH could be detected around term-equivalent age with the performed sequences.