Frequency and etiology of imaging diagnosis disagreements in children with prenatally diagnosed ventriculomegaly

Abstract

Purpose

To assess the frequency and etiology of variability in diagnoses on cranial ultrasound (US) and magnetic resonance (MR) imaging for children referred for prenatally diagnosed ventriculomegaly (VM).

Materials and Methods

Between 9/19/03-3/16/07, 119 children with US and MR studies performed within 13 months (median 6 days) after birth, after prenatal referral for VM, were studied as part of a prospective IRB-approved HIPAA-compliant study with written parental consent. 3 sonologists and 3 pediatric neuroradiologists interpreted the US and MR examinations, blinded to prenatal diagnosis. Final diagnosis was obtained by consensus (97 US, 53 MR and 31 US/MR comparisons). Ventricular size, types of disagreements, and reasons for disagreements were recorded. Disagreements on a per patient basis were categorized as major when they crossed diagnostic categories and had potential to change patient counseling.

Results

There was prospective agreement on 42/97 (43%) US and on 9/53 (17%) MR readings. Prospective consensus was more likely when the number of CNS anomalies was lower (P<.001 and =.002 for US and MR, respectively). In 24/55 (44%) of US and 11/44 (25%) MR with disagreements, one of the disagreements concerned the presence of VM. In 22/97 (23%) US studies and 22/53 (42%) MR studies the disagreements were potentially important. Reasons for discrepancies in reporting of major findings included errors of observation as well as modality differences in depiction of abnormalities. In comparing prenatal to postnatal diagnoses, there were 11/97 (11%) US and 27/53 (51%) MR examinations with newly-detected major findings, the most common being migrational abnormalities, callosal dysgenesis/destruction and interval development of hemorrhage.

Conclusion

Variability in postnatal CNS diagnosis is common after a prenatal diagnosis of VM. This is due in part to a lack of standardization in definition of postnatal VM.

Introduction

Ventriculomegaly (VM), a frequent fetal central nervous system (CNS) finding, is a common end-point for a variety of fetal pathologic processes. Determining an accurate prenatal CNS diagnosis is important for appropriate patient counseling and management. In order to assess the accuracy of prenatal diagnosis it is important to have a reliable reference standard for comparison.

One would expect postnatal imaging to serve as the reference standard for final CNS diagnosis in children with a fetal diagnosis of VM. However, the few published studies of postnatal imaging of children with fetal diagnosis of VM were limited by retrospective review ^1–4 or single reader interpretations of the postnatal imaging^5–8. To our knowledge, no large scale prospective studies have been performed to evaluate the accuracy of postnatal imaging by assessing inter-observer variability and how postnatal imaging diagnosis compares to fetal imaging diagnosis. Our study was performed to assess the frequency and etiology of variability in diagnoses on cranial US and MR imaging in children referred for prenatally diagnosed VM.

Materials and Methods

Subjects and Imaging

The study was performed at Beth Israel Deaconess Medical Center and Children’s Hospital, Boston as part of an Institutional Review Board approved, HIPAA compliant study evaluating fetal VM using US and MRI. Written informed consent was obtained from pregnant women who either had a referral diagnosis of VM or whom during an ultrasound examination were found to have VM (defined as a ventricular size measured on an axial view at the level of the atria greater than or equal to 10 mm). Prenatal US and MR examinations were performed and consensus imaging diagnoses were obtained on 195 women with 199 fetuses recruited from 7/1/03 – 8/20/06, as previously described⁹. Three were excluded after review of records showed they never had VM. Gestational age by ultrasound ranged from 16–41 weeks, with a mean of 26 weeks.

The imaging studies were grouped into 5 prenatal diagnostic categories to assess for attrition and bias in postnatal follow-up imaging: 1) normal prenatal imaging; 2) mild isolated VM (10–12 mm); 3) isolated VM 13–15mm; 4) isolated VM >15 mm; and 5) VM with other CNS anomalies (Table 1). For pregnancies resulting in a live delivery, the first postnatal head ultrasound (HUS) and/or brain MR examination were retrieved. If imaging was not performed for a clinical indication in the first 3 months of life, it was offered as part of the research study. However, many parents who felt their child was normal did not elect to have postnatal imaging. Postnatal imaging performed greater than 13 months of age were excluded from analysis. Our final population was 119 infants with postnatal imaging studies performed between 9/19/03-3/16/07. There were significantly fewer live births in the group with VM with other anomalies (p<0.0001, Table 1) compared to other diagnostic groups. There were significantly more postnatal MR examinations in the groups with >12 mm isolated VM and VM with other CNS anomalies groups than in the other diagnostic groups (p<0.001).

Table 1.

Prenatal diagnoses of subjects and enrollment in postnatal imaging portion of study

	N (%)
	Normal	Isolated VM 10–12 mm	Isolated VM 13–15 mm	Isolated VM >15 mm	VM with other CNS findings	All
Prenatal imaging†	21	85	15	2	73	196
Termination, stillbirth, neonatal demise	0 (0)	7 (8)	2 (13)	1 (50)	27 (37)	37 (19)
Liveborn, surviving neonatal period, able to be imaged postnatally	21 (100)	78 (92)	13 (87)	1 (50)	46 (63)	159 (81)
Liveborn, with potential for postnatal imaging†	21	78	13	1	46	159
Not performed before age 13 mo or images not available for review	9 (43)	26 (33)	2 (15)	0 (0)	3 (7)	40 (25)
Performed	12 (57)	52 (67)	11 (85)	1 (100)	43 (93)	119 (75)
Postnatal imaging †	12	52	11	1	43	119
Head ultrasound	10 (83)	50 (96)	8 (73)	1 (100)	28 (65)	97 (82)
MRI	3 (25)	6 (12)	5 (45)	1 (100)	38 (88)	53 (45)
Both	1 (8)	4 (8)	2 (18)	1 (100)	23 (53)	31 (26)

^†Normal and less severe diagnoses were associated with greater potential for liveborn, surviving the neonatal period, able to be imaged postnatally than were those with more severe diagnoses. Among potential subjects for postnatal imaging, normal and less severe diagnoses were associated with lower likelihood of actual performance of imaging; and among those imaged, higher likelihood of ultrasound and lower likelihood of MRI, compared with more severe diagnoses. All associations significant at p<0.001.

The postnatal HUS exams were performed from day of life 1 to 255 with a median age of 2 days after birth. MRI exams were performed from day of life 1 to 388 with a median age of 17 days after birth. Sonograms were performed on 61 male and 36 female children. MR examinations were performed on 29 male and 24 female children. The distribution of age at imaging did not differ by sex for either US or MR.

Figure 1 illustrates the overall study design. Postnatal neurodevelopmental follow-up of this cohort is described in a separate manuscript.¹⁰

Figure 1.

Consensus study design.

Sonogram performance, interpretation, and consensus

Prenatal sonograms were performed according to AIUM guidelines. Additional views of the head were obtained transvaginally when the fetus was in cephalic position. Sonograms were interpreted by one of 4 ultrasonologists (initials withheld for review, with 12–21 years experience in ultrasound) involved in patient care the day of the examination as well as by two blinded ultrasonologists, who only knew of the referral diagnosis for VM. VM was diagnosed when the ventricles at the level of the atrium measured ≥ 10 mm. Note that although the entry criteria for this study was referral for VM or the finding at time of sonography of VM (for a referral for some other indication), some fetuses at the time of confirmatory sonography were felt not have VM, and thus were coded as normal. CNS abnormalities were recorded using a modification of a coding system described by Van der Knaap et al.¹¹ The consensus diagnosis of prenatal subjects has been previously published⁹.

Postnatal HUS were performed according to institutional protocols and interpreted by 3 ultrasonologists (DL, CB, JE) who were blinded to prenatal diagnosis and clinical postnatal imaging diagnosis. The size of the ventricles at the atrium was measured. Unlike the prenatal imaging diagnosis of VM, where a measurement of 10 mm at the level of the atrium was used to define VM, no specific threshold was used for postnatal imaging, since, to our knowledge, none has been established. Reviewers therefore used their standard clinical practice subjective impression to diagnosis VM. Two neonates with holoprosencephaly were excluded from the ventricular measurement analysis due to inability to accurately measure the ventricles.

The readers rated their confidence in their diagnosis with regards to presence, character, and specific nature of the abnormality on a 5 point scale (1 = very confident to 5 = not confident).

The US diagnoses of the three ultrasonologists were compared for differences in opinion. Each disagreement was recorded as specified in appendix table 1. Disagreements of “no clinical difference” were those where similar abnormalities were being described, but with different codes, for example, codes for agenesis of the septum pellucidum and defect of the septum pellucidum. Errors of omission were those where by the description of the finding, the reviewer clearly saw the abnormality, but did not code for it. Errors of observation were those where neither description nor coding recorded the finding.

Appendix Table 1.

Reasons for disagreements and impact of disagreements used in this study

Reasons for disagreements (scored once for each disagreement)
	error of observation
	error of interpretation
	error of omission
	coding issue (two similar diagnoses have separate codes on the scale for example septo-optic dysplasia versus defect of the septum pellucidum versus agenesis of the septum pellucidum)
	disagreement regarding observation (still do not agree after consensus conference)*
	disagreement regarding interpretation (finding was seen by all in consensus conference, but no agreement on interpretation)*
	expected to be seen better at US (such as the wall of an arachnoid cyst)*
	expected to be seen better at MRI (i.e., parenchymal changes) *
	neuroradiologist experience would aid in diagnosis*
Reasons for a difference in opinion and the impact of the disagreements on the case (score as many as indicated for each case)
	decision to diagnose VM (difference of opinion as to whether VM was present)
	no clinical difference due to disagreement (agenesis of the corpus callosum versus damage of the corpus callosum)
	minor new finding
	major new finding
	overcall of a minor finding
	overcall of a major finding
	neonatal brain now appears normal*
Reasons for disagreement between prenatal and postnatal imaging
	resolution of VM
	worsening of VM
	ventricular size similar to prenatal examination but not coded as VM postnatally
	intervention (i.e., surgery or shunt placement);
	cortical migrational abnormalities more apparent at a later gestational age/postnatally;
	new CNS abnormality developed later (i.e., hemorrhage, tumor, or porencephaly);
	prenatal CNS abnormality resolved postnatally (i.e., hemorrhage);
	spinal neural tube defect not visualized on brain imaging
	corpus callosum dysgenesis more apparent
	coding issue
	inadequate view of the posterior fossa on postnatal HUS
	error
	abnormality of the type not expected to be seen on the imaging modality utilized (i.e., cortical migrational abnormality detected prenatally may be missed postnatally if only a HUS was done after birth and no brain MRI)
	other

^*Used only for prenatal to postnatal comparisons

Disagreements due to issues other than coding were settled by majority opinion of three prenatal sonologists at a consensus conference with image review. For a given child, all disagreements were evaluated to determine what potential impact the disagreement(s) had on each case. Major disagreements were those felt to be potentially clinically important, that could change patient counseling, as determined by our referring maternal fetal medicine guidelines, utilized in prior publications^{9, 12–14}. For example, a disagreement about the presence of a cyst in a fetus with agreement about the presence of agenesis of the corpus callosum was a minor change in diagnosis, but not clinically important. However, a disagreement about the diagnosis of dysgenesis of the corpus callosum when there was agreement about a midline cyst was a major new finding, with a potentially clinically important difference of opinion. The reason for a difference in opinion and the impact of the disagreements on the case are summarized in appendix table 1.

MR imaging performance, interpretation, and consensus

Fetal MR examinations were performed at 1.5 T typically using an 8 channel surface coil (in rare occasions with large patients in the third trimester, the body coil was utilized if a surface coil would not fit in the scanner). Sequence parameters varied during the study but always included single shot fast spin echo sequences in the fetal sagittal, coronal, and axial planes with 3–5 mm slice thickness, depending on gestational age and maternal body habitus. A typical sequence had echo spacing 4.2 msec, TE_effective 60 msec, matrix of 128 × 256, flip angle of 130 degrees and echotrain length of 72. Breathhold T1 weighted sequences were obtained in one or two fetal planes. A typical T1 sequence was turbo fast low angle shot technique with TR/TE of 15.4/4.2 ms, matrix 160 × 256, and slice thickness of 5 mm. Field of view for each sequence was tailored to the fetus and maternal body habitus. Other imaging sequences were performed at the discretion of the radiologist supervising the examination. Prenatal fetal MR examinations were interpreted by the radiologist supervising the examination (typically the ultrasonologist who performed the prenatal scan (DL, TM, JE, or CB) and by 3 pediatric neuroradiologists (CR, TYP, RR). The neuroradiologists were initially blinded to sonographic diagnosis and then re-interpreted the studies after knowledge of sonographic findings. Coded CNS abnormalities, ventricular measurements, and confidence ratings were recorded in the same manner described for US.

Postnatal MR examinations were performed according to institutional guidelines at 1.5T, typically with an 8 channel phased array brain coil and the following parameters: sagittal and axial SE T1 weighted MR images (TR/TE 450–550/9–14 msec; flip angle 90 degrees; 1 excitation; field of view 20× 24 cm; matrix 224 × 256; slice thickness 3.5 mm/skip 1 mm) and FSE T2 weighted images (TR/TE 2800 – 5000/98–108 msec; echo train length 16; 1 excitation, field of view 20× 24 cm, matrix 256 × 320 cm; slice thickness 4 mm/skip 1 mm). Additional sequences obtained on some neonates included coronal FSE T2, axial diffusion weighted images (B value of 1000, TR/TE 100000/85 msec; field of view 24 × 24 cm; matrix 128 × 128; slice thickness 4 mm skip 1 mm) and a susceptibility sequence (TR/TE 567/40 msec; flip angle 30 degrees; 1 excitation; field of view 20 × 24 cm; matrix 128 × 245; slice thickness 4 mm skip 1 mm). Neonatal images were typically obtained after feeding and wrapping the infant, without sedation. 50/53 (94%) neonatal MRs were performed without intravenous contrast and 3/53 (6%) were performed with contrast.

Postnatal brain MRI exams were independently reviewed by the same 3 pediatric neuroradiologists who interpreted the prenatal studies. They performed this review at least 3 months after prenatal imaging and without knowledge of prenatal diagnosis. Coded CNS abnormalities, ventricular measurements, and confidence ratings were recorded in the same manner described for US.

MR findings were compared for differences in coded diagnoses. Final postnatal MR diagnosis consensus was achieved by majority opinion of the pediatric neuroradiologists during a second image review session. The coding of disagreements as well as the etiology and/or impact of the disagreement were recorded in the same manner described for US.

Comparison of US and MR

A final postnatal consensus diagnosis was determined for each case. For subjects who had either an US or MR (and not both), the consensus diagnosis from that particular imaging modality was used as the final diagnosis. For neonates where both US and MR were performed postnatally, the US and MR consensus diagnoses were compared to one another. When the final diagnosis was ambiguous, one of the ultrasonologists (DL) with consultation from one of the pediatric neuroradiologists (CR) determined the final postnatal diagnosis. Studies with disagreements were coded as shown in appendix table 1. The etiology and/or impact of all the disagreements for each case were recorded in the same manner as for ultrasound.

Comparison of prenatal to postnatal imaging

Prenatal US, MR, and final diagnoses determined from our companion study.⁹ were compared to postnatal US, MR and final consensus diagnoses. Final prenatal to postnatal consensus diagnosis was performed at a separate image review session with two of the authors (DL, CR). For subjects with changes in diagnosis between prenatal and postnatal diagnoses, an explanation for the variability as well as the etiology or impact of the difference in diagnosis were tallied as described in appendix table 1. In cases where ventricular shunts had been placed and there was disagreement between the studies on the presence of VM, the assessment of VM was taken from the imaging study where the shunt was not present. For studies where VM was coded as being present prenatally but not present postnatally, the ventricular diameters were compared to assess if the measurement at the ventricular atrium had decreased or if it had remained at a similar level but was not coded as VM postnatally.

Statistical Analysis

Postnatal measurements of ventricular diameter were compared across categories of rater agreement by mixed-model analysis of variance (ANOVA), adjusting for within-subject and within-rater correlation. Where US and MR measurements were analyzed together, the ANOVA was also adjusted for imaging mode. Comparing studies with consensus with studies with disagreement, the rater’s confidence scores were analyzed by mixed-model ANOVA (adjusting for within-rater correlation) and the age at imaging and number of final diagnoses (including normal as a diagnosis) per patient by the Wilcoxon two-sample test to allow for the skewed distribution of those variables.

The number of children with structural abnormalities not visualized prenatally (choroid plexus cysts and hemorrhage were not included as structural abnormalities) were compared between different prenatal VM groups using Fisher’s exact test. Impact of gestational age at prenatal imaging, interval between prenatal and postnatal imaging, and age at postnatal imaging, and ventricular diameter (median of 3 measurements obtained at prenatal US) were individually assessed by Poisson regression analysis for association with the number of disagreements between pre-and postnatal imaging. SAS software version 9.1 (Cary, NC) was used for all computations. A p value of .05 was taken for statistical significance.

Results

For those fetuses with postnatal imaging, gestational age at prenatal ultrasound ranged from 17–41 weeks, with a mean of 27 weeks (Figure 2). Diagnoses are listed in Table 2 for the prenatal and postnatal imaging. In the final postnatal consensus 42/119 (35%) brains were judged normal. VM was present among the final diagnoses in 70/119 children (59%).

Figure 2.

Histogram of weeks of gestational age when prenatal imaging was performed for the 119 subjects in the final study population.

Table 2.

Prenatal and postnatal diagnoses, by US and MR readers, separately and by consensus.

Diagnosis*	Pre-natal	Postnatal					Final Prenatal to postnatal consensus
		US, any reader	US final consensus N (% of US, any reader)	MR, any reader	MR final consensus N (% of MR, any reader)	Final Postnatal US/MR consensus
Total subjects	119	97	97	53	53	119	119
VM	105	67	50 (75%)	48	45 (94%)	70	70
Normal	12	49	42 (86%)	3	2 (67%)	42	42
Dysgenesis corpus callosum	18	22	19 (86%)	23	22 (96%)	28	28
Hemorrhage	9	14	7 (50%)	14	12 (86%)	18	18
Cyst	6	9	9 (100%)	6	5 (83%)	12	12
Migrational abnormality/polymicrogyria	5	4	2 (50%)	13	11 (85%)	12	12
Chiari malformation**	8	6	5 (83%)	7	7 (100%)	9	9**
Porencephaly	7	6	3 (50%)	6	5 (83%)	5	7
Heterotopia	1	3	1 (33%)	6	6(100%)	6	7
Defect septi pellucidi	4	5	3 (60%)	8	7 (88%)	6	6
Congenital infarction	3	0	0	5***	5 (100%)	5	5
Dandy Walker variant/malformation	3	5	3 (60%)	6	4 (67%)	4	4
Cerebellar hypoplasia	2	5	1 (20%)	4	3 (75%)	3	3

^*Subjects may have more than one final diagnosis. This table lists only those diagnoses that occurred 5 or more times at any pointing the review process. For a listing of less frequent abnormalities, see appendix Table 2.

^**8 prenatal Chiari II diagnoses in fetuses with neural tube defects; 1 postnatal Chiari I malformation as a new finding

^***Includes one case of congenital infarction diagnosed at consensus conference

Ultrasound diagnoses and agreement

Of 97 children with postnatal HUS, there were 42 (43%) with prospective consensus (Table 3). Among these, 27/42 (64%) had normal final diagnosis, while the other 15/42 (33%) had 1–4 abnormal final diagnoses with a median of 1.

Table 3.

Disagreement in postnatal assessment, US-MR reconciliation, and prenatal-postnatal reconciliation.

	Postnatal assessment			Prenatal-postnatal reconciliation
	US	MR	US-MR	US	MR	Final
Total subjects	97	53	31	97	53	119
No disagreement, N (%)	42 (43%)	9 (17%)	4 (13%)	26 (27%)	13 (25%)	35 (29%)
At least one disagreement N (%)	55 (57%)	44 (83%)	27 (87%)	71 (73%)	40 (75%)	84 (71%)
Number of disagreements median, (range)*	2 (1–6)	3 (1–7)	3 (1–7)	1 (1–6)	2 (1–6)	2 (1–6)
Types of disagreement: N (%)†
(a) Decision to diagnose VM	24 (44)	11 (25)	3 (11)	25 (35)	0 (0)	22 (26)
(b) No clinical difference	23 (42)	27 (61)	5 (19)	16 (23)	4 (10)	10 (12)
(a) or (b)	44 (80)	30 (68)	8 (30)	41 (58)	4 (10)	32 (38)
(c) Major new finding	13 (24)	11 (25)	21 (78)	11 (15)	27 (68)	31 (37)
(d) Overcall, major finding	5 (9)	1 (2)	2 (7)	2 (3)	2 (5)	2 (2)
(e) Minor diagnostic change	12 (22)	11 (25)	1 (4)	5 (7)	10 (25)	9 (11)
(c) or (d)	16 (29)	12 (27)	22 (81)	13 (18)	28 (70)	32 (38)
Any of (c)–(e)	22 (40)	22 (50)	23 (85)	18 (25)	37 (93)	40 (48)
(f) Now normal	—	—	—	13 (18)	1(3)	14 (17)
Number of final diagnoses per patient
All, median (range)	1 (1–4)	3 (1–6)	3 (1–7)	1 (1–5)	3 (1–6)	1 (1–7)
No disagreement, median (range)	1 (1–2)	2 (1–4)	2 (1–2)	1 (1–3)	2 (1–3)	1 (1–3)
At least one disagreement, median (range)	2 (1–4)	3 (1–6)	4 (2–7)	1 (1–5)	3 (1–6)	2 (1–7)
p‡	<0.001	0.013	0.007	0.002	<0.001	<0.001

^*among those with at least one disagreement.

^†types are not mutually exclusive.

^‡p compares those with and without disagreement.

Of 55 neonates without prospective consensus, 15 (27%) had normal final diagnosis, and the other 40 (73%) had a range of 1–4 final abnormal diagnoses, with a median of 2. The number of final diagnoses was significantly greater in children without consensus (p<0.001).

In 24/55 (44%) sonograms with disagreements, one of the disagreements concerned the presence of VM (Figure 3). Measurements of ventricular diameter varied significantly according to whether the sonographers agreed on the presence of VM (Table 4, p<0.01).

Figure 3.

Coronal neonatal head ultrasound with the right ventricle measuring 9–11 mm and the left ventricle measuring 11–14 mm, diagnosed as normal by two sonologists and as ventriculomegaly by one sonologist.

Table 4.

Postnatal measurements of ventricular diameter, by imaging mode and prospective agreement or disagreement concerning presence of ventriculomegaly.

Mode	Agreement*	Examinations†	Measurements	Ventricular diameter, mm‡	Ventricular diameter, range
US
	All	95	270	12.9 ± 0.5	1–49
	Raters agree, VM present	41	118	18.3 ± 1.0	5–49
	Raters agree, VM absent	30	84	7.6 ± 1.2	1–11
	Raters disagree	24	68	10.2 ± 1.3	3–16
	• Rater indicates VM present	—	34	11.0 ± 1.3	6–15
	• Rater indicates VM absent	—	34	9.4 ± 1.3	3–16
	• Difference			1.6 ± 0.6 (p<0.01)
MR
	All	51	153	17.9 ± 0.6	3–46
	Raters agree, VM present	34	102	20.2 ± 1.2	11–46
	Raters agree, VM absent	5	15	8.5 ± 3.0	5–13
	Raters disagree	12	36	15.2 ± 2.0	3–30
	• Rater indicates VM present	—	27	15.5 ± 2.0	3–30
	• Rater indicates VM absent	—	9	14.3 ± 2.2	4–19
	• Difference			1.2 ± 1.2 (p=0.29)
US-MR comparison
	All	29	170	19.4 ± 0.7	3–46
	US/MR consensus, VM present	24	140	20.3 ± 1.4	4–46
	US/MR consensus, VM absent	2	12	9.8 ± 4.7	7–13
	US/MR consensus disagree	3	18	18.7 ± 3.8	3–46
	• Rater indicates VM present	—	11	24.0 ± 3.9	3–46
	• Rater indicates VM absent	—	7	10.6 ± 4.1	8–12
	• Difference			13.3 ± 2.6 (p<0.01)

^*Cases of agreement include those where raters disagreed concerning diagnoses other than VM.

^†Number of infants, each measured by 3 raters. Two instances of holoprosencephaly are excluded.

^‡Mean or difference ± standard error from analysis of variance, adjusted for within-infant and within-rater correlation. US-MR results also adjusted for imaging mode. Mean diameter varied significantly across categories of rater agreement, p<0.002, for each mode.

In the majority of studies with sonographic disagreements (44/55, 80%), at least one of the disagreements was a difference in opinion about diagnosing VM (N = 24) and/or had no clinical importance (N=23, Table 3). There were 22/55 (40%) subjects with potentially important disagreements. There were 114 disagreements on specific CNS diagnoses in the 55 subjects (Table 5). The most common type of disagreement was error of observation (N=45/114, 39%).

Table 5.

Disagreement on particular diagnoses in postnatal assessment, US-MR reconciliation, and prenatal-postnatal reconciliation

Type of disagreement*	US	MR	US-MR
Postnatal assessment†	114	139	98
Error of observation	45 (38)	24 (17)	0 (0)
Error of interpretation	16 (14)	25 (18)	4 (4)
Error of omission	7 (6)	13 (9)	1 (1)
Coding issue	25 (22)	77 (55)	23 (23)
Disagreement regarding observation	12 (10)	0 (0)	—
Disagreement regarding interpretation	9 (8)	1 (1)	—
Neuroradiologist experience would have helped	—	—	2 (2)
Expect to see better on US	—	—	11 (11)
Expect to see better on MR	—	—	48 (48)
Other	0 (0)	0 (0)	10 (10)
Prenatal-postnatal reconciliation	109	107	175
VM resolved	28 (25)	4 (4)	27 (15)
VM worsened	2 (2)	2 (2)	4 (2)
VM size unchanged, not coded as VM postnatally	11 (10)	1 (1)	12 (7)
Cortical migration more apparent	3 (3)	18 (17)	19 (11)
New CNS abnormality developed later	15 (13)	33 (30)	38 (22)
Prenatal abnormality resolved	3 (3)	5 (5)	8 (5)
Callosal dysgenesis now apparent	10 (9)	4 (4)	11 (6)
Coding issue	17 (15)	20 (18)	21 (12)
Inadequate view posterior fossa	1 (1)	0 (0)	0 (0)
Errors	5 (4)	2 (2)	7 (4)
Other	10 (9)	18 (17)	26 (15)

^*Number (%) of diagnoses on which raters disagreed entering consensus conference; may include more than one per patient.

^†Types of disagreement are not mutually exclusive in postnatal assessment.

The median age at postnatal US imaging was 2 days and did not differ between those with and without consensus (p>0.50).

In the 42 subjects with consensus, the sonologists’ confidence was higher with regard to the types of abnormality (p<0.05) than in the 55 subjects without consensus (Figure 4). The level of confidence in additional findings associated with the anomaly was generally not as high as that in the presence or nature of the anomaly itself, but was significantly higher in subjects with consensus than without (p<0.001).

Figure 4.

Confidence of US and MR readers in postnatal diagnosis of fetal CNS abnormalities. Bars indicate mean ± standard error on a five-point scale from very confident to not confident. P-values from analysis of variance, adjusted for inter-reader and inter-subject variability.

MR diagnoses and agreement

Of the 53 postnatal MRI examinations, consensus was reached prospectively by 3 pediatric neuroradiologists in 9 infants (17%, Table 3). Two subjects were normal, while the other 7 had 1–4 abnormal final diagnoses, with a median of 2. There were 44/53 subjects (83%) without consensus. These had a range of 1–6 abnormal final diagnoses, with a median of 3. The number of final diagnoses was significantly greater in subjects without consensus (p=0.013).

In 11/44(25%) of MR subjects without consensus, there was disagreement regarding the presence of VM (Table 3). Ventricular diameter did not differ significantly between raters indicating VM present and those indicating VM absent (Table 4).

The majority of disagreements on MR (30/44 = 68%) either had no clinical importance (N=27) and/or involved a disagreement about whether or not to diagnose VM (N=11, Table 3). In 22/44 instances (50%), the disagreements were categorized as being potentially important.

On MR exams there were 139 disagreements on specific CNS diagnoses (Table 5). The most common types of disagreements were coding issues (N=77, 55%). The neuroradiologists’ level of confidence in MR findings with respect to the presence and nature of abnormalities and associated findings was generally higher than that for US readings (Figure 4). Confidence in MR findings did not differ between subjects with or without consensus.

The median age at postnatal MR imaging was not significantly greater in children of consensus (21 days, range 1–388 days) than in those with disagreement (10 days, range 1–383) (p>0.40).

Comparing US and MR postnatal interpretations

There were 31 subjects where both a postnatal HUS and MRI were performed. There were 27 subjects without consensus (87%, Table 3). These had a range of 2–7 final diagnoses, with a median of 4. The number of final diagnoses was significantly greater in subjects without consensus (p=0.007).

Among the 27 HUS/MR examination pairs with disagreements, the disagreement concerned calling VM (N=3) or had no clinical importance (N=5); one or both of these accounted entirely for the difference of opinion in 4 instances (15%). In the remaining 23 studies there were 21 major new findings, two overcalls of a major finding, and 1 minor change in diagnosis. The common findings seen on postnatal MR which were not recognized on postnatal HUS were migrational abnormalities (N=8), hemorrhage (N=7), and infarction (N=4).

Comparison of Prenatal to Postnatal imaging

Prenatal and postnatal final diagnoses agreed in 35/119 studies (29%, Table 3). The majority of these (30/35, 86%) were either normal (N=9) or isolated VM (N=21). In the remaining 84 studies (71%) there was initial difference in diagnostic coding. The group with consensus had fewer diagnoses than the group without consensus (p<0.001).

The range of interval between prenatal and postnatal US imaging was 1–338 (mean 106, median 100) days. In US imaging (prenatal vs. postnatal) there was agreement in 26/97 studies (27%), with the majority of these having final CNS diagnosis of either isolated VM (N=16) or normal (N=8). In 71/97 (73%) studies there were differences in diagnoses and the final diagnoses varied from normal (N=34) up to 5 coded CNS abnormalities, most commonly dysgenesis of the corpus callosum (N=16).

The range of interval between prenatal and postnatal MR imaging was 5–455 (mean 129, median 98) days. In MR imaging (prenatal vs. postnatal) there was agreement in 13/53 studies (25%). In 5 of those studies (38%) the final MR diagnosis was either normal (N=1) or isolated VM (N=4). There were 40/53 studies (75%) with differences in final diagnosis (Figure 5–8). In 20/40 (50%) of those studies with there were 3 or more coded abnormalities, the most frequent being dysgenesis of the corpus callosum (N=15) and polymicrogyria (N=9).

Figure 5.

Hemimegalencephaly misdiagnosed as hemorrhage on prenatal imaging. A. Axial single shot fast spin echo (SSFSE) T2WI at 22 weeks gestational age shows some areas of low signal intensity (arrowheads) felt to represent subependymal hemorrhage with dilatation of the right lateral ventricle at initial prenatal interpretation. The region of asymmetric enlargement and irregularity of the cortex (arrows) and generalized enlargement of the right cerebral hemisphere was not prospectively noted. B. Axial fast spin echo (FSE) T2WI on day 1 of life shows a mildly larger right cerebral hemisphere and polymicrogyria that is most marked in the right frontal lobe (arrows). There is abnormal hypointense signal within the right frontal white matter and basal ganglia. These findings are consistent with hemimegalencephaly.

Figure 8.

Example of new development of an abnormality and error of observation. A. Sagittal (A) and axial (B) SSFSE T2 weighted MR at 31 weeks gestational age show the corpus callosum (arrowhead on A, short black arrow in B). Note also absent leaflets of the septum pellucidum (*) and left temporal schizencephaly (long arrow) with ventriculomegaly. C. Sagittal SE T1-weighted MR at 3 months of age shows deficiency in the anterior genu and rostrum of the corpus callosum (arrows) that was not appreciated on the prenatal MR. Note normal appearing body of corpus callosum, as seen prenatally (arrowhead). D,E. Axial FSE T2-weighted MR at 3 months of age show absent septal leaflets, deficiency of the anterior genu of the corpus callosum (black arrowhead) with associated dysmorphism of the frontal horns of the lateral ventricles and schizencephaly (long arrow). In addition there are subependymal neuronal heterotopia (white arrowheads) not visible on the prenatal study and not recorded by one of the neuroradiologists.

In 38/53 (72%) studies, the postnatal MR diagnoses matched the final diagnoses. An example of a lesion present, but not noted on prenatal imaging, is shown in figure 6. However, in one instance the prenatal diagnosis was felt to be more accurate, i.e., the postnatal MRI suggested the diagnosis of schizencephaly whereas the fetal MRI diagnosis was encephaloclastic porencephaly (Figure 7). In this case, the postnatal diagnosis was schizencephaly since cortex seemed to line the defect. However, since by prenatal imaging this was in the process of developing, it was felt to be encephaloclastic event, rather than a genetic event.

Figure 6.

Congenital CNS tumor as an error of observation. A. Sagittal SSFSE T2 weighted MRI at 36 weeks gestational age showed VM and a very subtle area of low signal intensity above the tectum (arrow). This was only noted in retrospect after postnatal MRI. B. Postnatal spin echo (SE) T1WI without contrast shows a mildy hyperintense mass above the tectum (arrow). This was felt to be a hamartoma or low grade glioma causing hydrocephalus due to obstruction at the level of the aqueduct.

Figure 7.

Example of prenatal diagnosis being more accurate than postnatal diagnosis. A. Coronal SSFSE T2WI at 33 weeks gestational age shows ventriculomegaly and a region of porencephaly with slightly higher signal intensity fluid (arrow). B. Axial fast spin echo (FSE) T2 weighted MRI on day of life 27 (with a 2 month interval from fetal MRI) shows the extra-axial fluid appears contiguous with the ventricular system (arrow). Initial interpretation of the postnatal MR (blinded to prenatal diagnosis) included schizencephaly since the parenchyma appears to have a cortical rim (arrow) in the region of the defect. However, when interpreted in conjunction with the fetal MRI, the finding was felt to represent porencephaly. Additional findings on this image are dysmorphic ventriculomegaly with absence of the septum pellucidum and a large extra-axial fluid collection with midline shift. The patient had other features (not shown) consistent with lobar holoprosencephaly.

Low signal intensity lesions were a relatively common cause of difference in final diagnosis. For example, nodular subependymal T2 shortening could be due to germinal matrix hemorrhage or neuronal heterotopia in a fetus with hemimegalencephaly. The case of hemimegalencephaly was interpreted on prenatal MR as germinal matrix hemorrhage with dilatation of the ipsilateral ventricle as a result of hemorrhage. More extensive associated cortical malformation and the correct diagnosis only became apparent as the brain matured (Figure 5).

Ventricular diameter on prenatal imaging was significantly associated with the number of disagreements needing reconciliation in the final prenatal-postnatal consensus conference (p<0.001). Median diameter ranged from 10.5 mm in those with consensus in the final conference (n=36) to 24 mm in those with 5–6 disagreements. The number of disagreements was not associated with gestational age at prenatal imaging, interval between prenatal and postnatal imaging, or age at postnatal imaging.

There were 12 fetuses categorized as normal prenatally. In 10 of these postnatal diagnosis was normal. Two were felt to have VM as neonates, one with a small choroid plexus cyst (Table 6). Of 52 fetuses with isolated mild VM 10–12 mm prenatally, 28 had normal postnatal imaging, 17 had isolated VM, 3 had intracranial hemorrhage, 2 had small choroid plexus cysts and 2 (4%) had structural abnormalities not diagnosed prenatally (Table 6). Of 11 fetuses with isolated VM 13–15 mm prenatally, 2 had normal postnatal imaging, 3 had isolated VM, and 4 (37%) had structural abnormalities not diagnosed prenatally. 1 fetus with VM >15 mm prenatally was diagnosed with a tumor postnatally. In these three groups without additional abnormalities prenatally, there were increasing numbers of structural abnormalities seen postnatally as the degree of VM increased (p<.0001).

Table 6.

Postnatal diagnoses with respect to prenatal diagnostic group

Prenatal diagnosis	Postnatal diagnosis, N (%)
	All	Normal	Isolated VM	VM with other abnormalities (N)
Normal	12	10 (83)	1 (8)	1 (8): Choroid plexus cyst (1)
Isolated VM, 10–12mm	52	28 (54)	17 (33)	7 (13): Hemorrhage (3) Choroid plexus cyst (2) Cyst with heterotopia (1) Abnormal midbrain with dysgenesis of corpus callosum and migrational abnormality (1)
Isolated VM, 13–15mm	11	2 (18)	3 (27)	6 (55): Defect of septum pellucidum (2) Chiari malformation (1) Hemorrhage and scalp mass (1) Hemorrhage and mega cisterna magna (1) Dysgenesis of corpus callosum (1)
Isolated VM, >15 mm	1	0 (0)	0 (0)	1 (100): Hemorrhage and tumor (1)
VM with other CNS findings	43	2 (5): Cerebellar hypoplasia (1) Mega cisterna magna (1)	0 (0)	41 (95)
All	119	42 (35)	21 (18)	56 (47)

In 37 fetuses VM was coded as being present prenatally, but not present postnatally. In 12/37 (32%) of these the ventricular atrium measured a similar amount prenatally compared to postnatally but these were not coded as VM in the postnatal study. In 25/37 (68%) the degree of ventricular dilatation decreased between the examinations.

Discussion

Disagreements are common among readers of postnatal imaging studies after a fetal diagnosis of VM, and are more likely as the degree of VM increases and as the complexity of CNS anomalies increases. In addition to differences of opinion and error of observation, there are a number of reasons for discrepancies between prenatal and postnatal imaging. These include interval resolution of VM (25/200, 12.5% of our population), differences in criteria for diagnosis of VM (12/200, 6%) ventricles measured a similar amount but were not coded as VM postnatally, new development of a CNS abnormality over time (i.e., porencephaly); certain abnormalities becoming more apparent at a later gestational age (i.e., a cortical migrational abnormality and dysgenesis of the corpus callosum), and resolution of an abnormality, such as hemorrhage. Importantly, additional MRI sequences (for example, susceptibility sequences showing hemorrhage that can be utilized in neonatal MRI but are not as easily adapted to the rapid imaging needed for fetal MRI) and contrast medium (for example in a fetus with a brain neoplasm) can be utilized postnatally, and not in fetal MRI, may allow for improved visualization of previously unsuspected abnormalities.

The diagnosis of VM was a frequent area of disagreement (25% and 21% in US and MR subjects, respectively). In the prenatal population, VM is clearly defined as atrium of the lateral ventricle measuring 10 mm. However, to our knowledge, there is no such definition for VM in the postnatal population. There are a wide variety of methods for characterizing neonatal ventricles using US (such as the ratio of the distance from the falx to the lateral wall of the ventricle to the hemispheric width, ratio of ventricular diameter to the diameter of the brain at the same level, displacement of the medial wall of the ventricle toward the midline, and subjective assessment)^15–26. Similarly, on neonatal MRI, ventricular size has been assessed using a ventricular/brain ratio²⁷. We found a significant difference of opinion (p<.01) between radiologists interpreting postnatal HUS as to when VM should be diagnosed postnatally. The prenatal definition cannot be utilized in postnatal HUS studies because imaging through the anterior fontanelle of the neonate does not give the sonologist the same axial plane of view. Even on postnatal MR with axial imaging planes, ventricles can measure larger than 10 mm and not be classified as enlarged. Therefore, without an interval change in the size of the ventricles, an individual could be given the diagnosis of VM in utero, but after delivery (even on the same day) be considered normal because of the lack of a standardized definition. However, the clinical importance of this difference in diagnosis is not yet established. Our companion paper¹⁰ assesses the outcome of fetuses with varying degrees of prenatally diagnosed VM in our population.

It is well recognized that VM can resolve in utero. In our study, of 12 fetuses referred for VM who were felt to be normal at prenatal imaging, 2 had postnatal studies showing VM. This suggests that even when VM resolves in utero, postnatal follow-up should be obtained. Structural abnormalities have been reported postnatally in up to 10% of subjects with isolated mild VM diagnosed prenatally.^28–30 We demonstrated that the risk of postnatal diagnosis of structural abnormalities increases as the degree of VM increases, with new structural abnormalities seen in 0% of fetuses judged to be normal prenatally, 4% of fetuses with VM 10–12 mm prenatally, and 37% of fetuses with VM 13–15 mm prenatally.

An important limitation of our study is the bias in the study population due to attrition of fetuses with complex CNS anomalies who did not survive to birth to undergo postnatal imaging. In addition, there were differences in performance and modality of imaging after birth that were driven by prenatal diagnosis. While the group with prenatally diagnosed VM with associated CNS anomalies had the largest prenatal attrition, they also had the highest percentage of postnatal imaging when liveborn (93%) and the highest likelihood of having a postnatal MRI (88%). Subjects in the normal or mild isolated VM groups often did not complete postnatal follow-up. In addition, there was a higher percentage of disagreement in the MR subjects (83%) compared to the US subjects (57%). Much of this discrepancy can be explained by the fact that the children with more abnormalities prenatally had postnatal MR examinations more often than those with fewer anomalies. This biases our results to studies with more diagnoses, and thus more discrepancies. Migrational abnormalities, hemorrhage, and infarctions were better visualized on postnatal MR compared to US. In children without MR imaging, such abnormalities could have been missed. Recall bias is a possibility, as readers overlapped between prenatal and postnatal imaging studies. However, the length of time between obtaining these studies should have decreased this potential bias. A final limitation is lack of correlation of our diagnoses with developmental outcomes. This is addressed in a prior publication from our study.¹⁰

Normal brain development continues throughout pregnancy and even postnatally, therefore it is often not possible prenatally to tell if a structure will be normal, merely that it appears normal for that stage in gestation. Furthermore, abnormalities detected antenatally can evolve, e.g. mild VM may become severe VM, and new abnormalities may evolve, for example a region of hemorrhage may become an area of porencephaly. Also, head size increases so there is improvement from the imaging perspective since larger structures in general are easier to evaluate. These issues will affect the comparison of prenatal to postnatal findings, and are thus important concepts to understand when counseling patients after a diagnosis of fetal VM. In this study we have assessed congenital findings that have of necessity changed in character and appearance over time. It is possible that some lesions were acquired after a postnatal event. However this type of information is also important to consider when counseling the parents.

In conclusion, disagreements among radiologists are common with regard to the final CNS diagnosis for children with a prenatal diagnosis of VM. This leads to difficulty in establishing a reference standard for accuracy of prenatal diagnosis. This is particularly problematic in the postnatal characterization of VM. Understanding the variability in postnatal diagnosis after a prenatal diagnosis of VM is important to clinicians who care for and counsel these patients. An imaging standard for postnatal diagnosis of VM is needed to improve consistency in reporting. However, it should be recognized that change in appearance over time, and clinical outcome are needed to assess the clinical importance of fetal VM.

Appendix table 2.

Uncommon abnormalities, prenatal and postnatal diagnoses, by US and MR readers, separately and by consensus.

Diagnosis*	Pre-natal	Postnatal					Final Prenatal to postnatal consensus
		US, any reader	US final consensus N (% of US, any reader)	MR, any reader	MR final consensus N (% of MR, any reader)	Final Postnatal US/MR consensus
Periventricular leukomalacia	1	1	0 (0%)	4*	(100%)	4	4
Congenital cerebral calcification	0	1	1 (100%)	1	1 (100%)	2	2
Holoprosencephaly	1	2	0 (0%)	2	2 (100%)	2	2
Megacisterna magna	3	0	0	2	1(50%)	2	2
Abnormal midbrain/thalamus	0	0	0	2	2(100%)	2	2
Craniosynostosis	1	0	0	2	2(100%)	2	2
Ectopic posterior pituitary	0	0	0	1	1(100%)	1	1
Schizencephaly	1	0	0	2	2(100%)	2	1
Micrencephaly	1	0	0	1	1(100%)	1	1
Tumor	0	0	0	1	1 (100%)	1	1
Scalp mass	0	0	0	1	1 (100%)	1	1
Hemimegalencephaly	0	0	0	1	1 (100%)	1	1

^*includes one case of PVL diagnosed at time of consensus conference.