MRI classification system (MRICS) for children with cerebral palsy: development, reliability, and recommendations

Surveillance of Cerebral Palsy in Europe (SCPE) has, through agreement in definitions and classifications of cerebral palsy (CP), developed a common language in the domain of CP.[1, 2] This has been a prerequisite for comparative studies,[3, 4] and for studies which necessitate a large population basis.[4, 5] This common language has generated a lot of interest even beyond Europe: the definitions and classifications are widely used, and SCPE papers extensively cited.

CP is a clinical diagnosis, based upon neurological symptoms and a motor disorder causing an activity limitation.[6] SCPE does not consider neuroimaging a prerequisite for the diagnosis of CP.[2] Up to now, normal magnetic resonance imaging (MRI) does not exclude the diagnosis of CP. Moreover, neuroimaging, especially MRI, is not available nor used in all countries to the same extent. In addition, the use of MRI, as well as knowledge of its role in the understanding of CP pathogenesis, has dramatically increased during the last 15 years, which would make comparison between countries and time periods difficult. Most importantly, there is no commonly agreed neuroimaging classification for CP at this time.

Although neuroimaging is not part of the CP definition, neuroimaging findings are abnormal in more than 80% of children with CP,[7-9] disclosing the pathogenic pattern responsible for the CP. Neuroimaging may also help to understand the structure–function relationship.[10, 11]

National guidelines recommend MRI as the first diagnostic step after history taking, neurological examination, and examination of additional impairments.[12] Therefore, a need was expressed for the development of a classification system for neuroimaging findings in CP, which could be introduced into the standard CP evaluation form to be prospectively used in registers. For this purpose, a simple system is needed, which can be applied easily not only by clinicians, but also by epidemiologists or other professionals who work in CP registers. In CP registers, in most cases only a written description of the imaging findings is available and not the image itself. The classification therefore has to be applicable also for the written descriptions.

With the aim to develop such a classification, SCPE had as a first step identified MRI patterns known to be associated with a high risk of CP. CP is defined as a disorder due to a non-progressive interference or lesion or abnormality of the developing or immature brain.[1] Thus, it seemed important to identify patterns according to their time of occurrence during brain development, on the basis that the human brain undergoes complex organizational changes during intra- and extrauterine development; the patterns of abnormalities or lesions found will depend on the stage of brain development at the time of insult.[13] Thus, the brain pathology of CP is dependent on the time of occurrence of noxious events interfering with brain development or actually damaging it. MRI has good potential to visualize this pathology. During the first and second trimesters, predominantly cortical neurogenesis takes place, characterized by proliferation, migration, and organisation of neuronal precursor cells, then neuronal cells. Disturbances in this process result in maldevelopments such as lissencephaly, pachygyria, or polymicrogyria. When affecting the motor cortex, CP may be the result. Some of these disorders may be of genetic origin, especially when symmetrical in distribution.[14]

During the third trimester, when the ‘gross architecture’ of the brain (neural cyto- and histogenesis) is largely established, growth and differentiation events are predominant and persist into postnatal life (axon, dendrite and synapse formation, myelination). Disturbances of brain development during this period mainly result in clastic lesions. The causes are multiple and key factors are inflammation with excessive cytokine production, oxidative stress, and excess release of glutamate triggering the excitotoxic cascade, factors that are induced by hypoxic–ischaemic and/or infectious mechanisms, whereby a potentiation of single effects can be assumed.[15] Early in the third trimester white matter is especially affected. The major neuropathology potentially damaging the motor tracts and thus leading to CP includes periventricular leukomalacia (PVL) or complications of intraventricular haemorrhage: predominant white matter injury. This pathology has been called ‘encephalopathy of prematurity’ and is accompanied by ‘neuronal/axonal disease’. This may also involve the thalamus, basal ganglia, cortex, brainstem, and cerebellum,[16] although the pathology that is related to CP and identifiable by visual inspection is mainly that of white matter injury.

Lesional patterns that can occur in central motor domains and thus cause CP in the late third trimester concern the cortical grey matter, basal ganglia, and thalamus:[17] predominant cortical or deep grey matter injury. Infarcts of the middle cerebral artery are reported mainly in children born at or near term with unilateral spastic CP.[18] They may also occur in the very-preterm child, then involving more the lenticulostriate arteries.[19]

Thus pathological patterns, characterizing disturbance of early brain development and likely to cause CP, can be summarized as maldevelopments, predominant white matter injury, and predominant grey matter injury; we called these patterns ‘pathogenic patterns’ because they characterize specific timing periods of disturbance/insult to the developing brain.

In a second step, we had systematically searched in the literature for these pathogenic patterns in children with CP.[7] In this earlier review, MRI was reported to be abnormal in 86% of the CP cases and indicated the pathogenesis in 83%. Periventricular white matter injury was the most frequent finding (56%), followed by deep grey matter injury (18%), and brain maldevelopment (9%). Thus, this earlier review indicated that the classification based on pathogenic patterns covers the main brain pathology in children with CP and seems a useful approach also for CP registers.

The aim of this study was to develop an MRI classification for CP research, registers, and clinicians, compare it with other classification systems used in the literature, and test its reliability.

Method

The development of a classification system was based on the concept of the underlying pathogenic MRI patterns in children with CP. Typical illustrations of these patterns including subgroups and descriptions in the context of clinical cases were searched for and agreed on by the authors. A manual was developed in the process and tested with respect to comprehensibility and applicability during several SCPE workshops. Type of CP and gross motor function according to Gross Motor Function Classification System (GMFCS)[20] were described together with MRI findings.

A literature review was performed, searching for other classification systems of MRI findings in CP, with the question of compatibility with the MRI classification system (MRICS), and whether harmonization between classifications was possible. This review covered the time period January 2007 to March 2013. The following keywords were used for the search: cerebral palsy, neuroimaging, MRI, classification.

A workshop with SCPE participants aimed at harmonizing the findings of the different classifications with the MRICS.

A reliability study of the suggested classification system was done using 18 MRIs and their written reports; these illustrated maldevelopments (3 cases), predominant white matter injuries (7), predominant grey matter injuries (5), miscellaneous (2), or normal findings (1). Age at examination and CP subtype were indicated. The test case examples were not part of the MRICS manual. These examples together with the manual were sent to all SCPE centres asking for participation of members involved in collecting neuroimaging results for the register. The participants were asked to do the study either with the images, or with the written reports, using the MRICS manual as reference. They were asked to classify at least the main pathogenic group (A, B, C, D, E), and in the case of more than one pattern, to classify the predominant pattern or the pattern that most probably caused CP. MRICS interrater reliability was estimated with kappa statistics (and 95% confidence intervals [95% CI]) and interpreted according to usual ordinal categories.

Results

The classification system based on the pathogenic patterns including subgroups is described in Table 1, and was called MRICS.

A manual exemplifying and illustrating these patterns, including clinical cases for training, was established (www.scpenetwork.eu/en/my-scpe/rtm/neuroimaging/cp-neuroimaging/). Examples and definitions – such as unilateral or bilateral involvement, or mild versus severe extent of a lesion – were included. All examples indicated the age at imaging, the CP subtype and severity according to GMFCS, and the topography of the lesion and its relation to the CP subtype were explained. The aspect of topography and functional consequence was highlighted in a specific chapter. Figure 1 gives an overview on brain development and timing in relation to the pathogenic patterns described. Figures 2 to 4 exemplify MRI findings in the A, B, and C categories respectively; in the B category, also with respect to morphology–function. In a second part of the manual, concrete case reports of children with CP integrated the MRI in the clinical context. Thus, the classification system is primarily qualitative (patterns of different timing). To some degree it is also quantitative, because parameters are introduced such as unilaterality or bilaterality, and for some patterns – such as PVL or diencephalic lesions – also aspects of lesion extent.

Specific pitfalls were highlighted, such as the importance of age of the child at MRI. Incomplete myelination may make it difficult or impossible to identify mild PVL or mild deep grey matter lesions.

The cases in the manual can be studied primarily as illustrations of specific patterns, or without revealing the MRI classification, so that the user can train themselves. This manual was included in the Reference and Training Manual of the SCPE (www.scpenetwork.eu/en/rtm/).[2]

The literature review for other classification systems of MRI findings in CP identified 49 publications, 10 of which qualified for further analysis because they used specific classification systems: two were reviews,[8, 9] five population based studies,[11, 21-24] and three focused on a specific CP subtype.[25-27] All these studies had in common a high proportion of abnormal MRIs (83–87%). Most of the items used in the different classifications could be easily allocated to one of the groups suggested in the SCPE classification system as they referred to comparable pathogenic patterns. Minor differences concerned the wording of an item such as maldevelopment versus malformation; harmonization thus seemed unproblematic. Major differences concerned patterns without a corresponding term in the SCPE classification. An important category here refers to enlargement of intra- and/or extracerebral spaces such as ventriculomegaly, atrophy, cerebrospinal space abnormalities, cerebellar atrophy, diffuse cortical atrophy, thin corpus callosum, and enlargement of the lateral ventricles. Other items without correspondence were infection and hypomyelination (Table 2).

Multiple lesions (e.g. more than one abnormality where it is difficult to decide which one is predominant or causing CP) were not dealt with unanimously and were generally considered to be problematic for classifying. The age at MRI varied considerably and there was no consensus on how to deal with the problem of incomplete myelination. There was a tendency however, to include MRI performed at a younger age and even neonatal MRI into the analysis.

Discussion

There is a consensus on an international basis that neuroimaging is of great importance in the diagnosis of CP.[7-9, 12] Neuroimaging is abnormal in more than 80% of children with CP. Neuroimaging helps understanding of pathogenesis and, to a minor extent, also aetiology of the underlying brain disorder. It may also help to understand the structure–function relationship.[10, 11] Thus a classification system for MRI findings in children with CP seems to be of high importance. Especially for CP registers, there is a great need for consensus on classification. In the literature, the need for a standardized classification system is expressed in every study dealing with MRI in children with CP. In most of the studies, some kind of classification is indeed used.

Brain abnormalities in CP arise at different times during brain development. The same cause, for example, an infectious agent or a hypoxic–ischaemic event, may give rise to different patterns depending on the timing of interference with brain development. Therefore, it seemed logical to describe pathogenic patterns related to timing rather than use a classification based on aetiology, especially because aetiology is often not clear.

After having agreed on definitions concerning CP and its subtypes as well as additional impairments, pre-, peri-, and neonatal data and denominators,[28] SCPE now suggests a classification system for the reporting of brain abnormalities identified by means of MRI in children with CP: the MRI classification system, MRICS. A manual developed under continual feedback of the SCPE group illustrates these brain abnormalities in three major groups and their subgroups, including morphology–function aspects, and can be used as a training tool. MRICS is primarily a qualitative system based on pathogenic patterns related to timing, but it does take into account different pathologies within one timing period, which are partly also related to the extent of a lesion. Thus, MRICS has some simple quantitative aspects (e.g. uni- vs bilaterality, severity of a pattern such as basal ganglia/thalamus lesions). More sophisticated quantitative systems, such as the scoring system for MRI in CP,[29] could be used in addition to describe the extent of lesions within the pathogenic groups. This system analyses the brain injury in children with CP semi-quantitatively, in scoring the extent of a lesion (the higher the score, the more extensive the lesion); information on topography is given when subscores are considered, but lesion type itself is not addressed. Thus, the systems appear complementary. When comparing MRICS to other classification systems used in the literature, harmonization seemed feasible because, usually, similar patterns were addressed, although with a somewhat different wording, and some aspects had to be allocated to the miscellaneous group (e.g. hypomyelination, cerebellar atrophy, or unspecified atrophy). So, in a next step, we would like to test MRICS also at an international level to see whether consensus on the classification can be achieved.

MRI findings are increasingly introduced into CP registers; usually the imaging report is used, but some registers have direct access to images. For both, rating images directly and ratings of imaging reports, we could show a very good interrater reliability.

Several questions turned up repeatedly in this context which led to the following recommendations noted in the SCPE guideline for data submission and in the Reference and Training Manual (www.scpenetwork.eu/en/my-scpe/rtm/neuroimaging/cp-neuroimaging/recommendation-for-performing-mri-in-cp/):

• If there are several patterns, the predominant pattern that is believed most likely to have led to the CP should be classified first; if there is another pathogenic pattern (A, B, or C), this should be classified separately.
• Patterns should be classified as a minimum at headline level (A, B, C, D, E); if feasible, patterns should be further classified at the subgroup level.
• Ideally, an MRI performed after the age of 2 years should be used for classification purposes, because of developing myelination during the first years. If MRI performed before 2 years of age is normal, patterns such as mild periventricular or basal ganglia/thalamus lesions may have been be missed; and MRI should be repeated. If a clear pathogenic pattern (A, B, C) is found before the age of 2 years, this MRI is appropriate for classification.

In a longer process involving clinicians and non-clinicians, SCPE has developed a classification system for MRI in CP, which proved easy to use and reliable. In comparison with other classification systems described in the literature, harmonization appeared feasible. MRICS is a mainly qualitative classification system for cerebral palsy describing pathogenic neuroimaging patterns related to timing of brain compromise. A web-based manual is available for training purposes.

Acknowledgements

This study was performed on behalf of the SCPE collaboration and was funded by the European Union Health Programme – grant numbers DG SANCO EAHC 2008–1307/2013–3211 – ‘Surveillance of Cerebral Palsy in Europe: best practice in monitoring, understanding of inequality and dissemination of knowledge’.

We wish to thank the members of the SCPE who in many ways have contributed to the development of the MRICS. List of SCPE participants: C Cans, M Van Bakel (RHEOP, Grenoble, France); C Arnaud, M Delobel (RHE31, Toulouse, France); J Chalmers (ISDSHS, Edinburgh, UK); V McManus, A Lyons (Lavanagh Centre, Cork, Ireland); J Parkes, H Dolk (Belfast, UK); K Himmelmann, M Pahlman (Göteborg University, Göteborg, Sweden); V Dowding (Dublin, Ireland); A Colver, L Pennington (University of Newcastle, Newcastle, UK); K Horridge (NECCPS, UK); J Kurinczuk, G Surman (NPEU, Oxford, UK); MJ Platt (University of Liverpool, Liverpool, UK); P Udall, G Rackauskaite (NIPH, Copenhagen, Denmark); MG Torrioli, M Marcelli (Lazio Cerebral Palsy Register, Rome, Italy); G Andersen, S Julsen-Hollung (CPRN, Tonsberg, Norway); M Bottos (Bologna, Italy); G Gaffney (Galway, Ireland); J De La Cruz, C Pallas (DIMAS-SAMID, Madrid, Spain); D Neubauer, M Jekovec-Vrhovšek (Ljubljana, Slovenia); D Virella, M Andrada (Lisbon, Portugal); A Greitane (Riga, Latvia); K Hollody (Pecs, Hungary); S Sigurdardottir, I Einarsson (Reykjavik, Iceland); M Honold, K Rostasy (Innsbruck, Austria); V Mejaski-Bosnjak (Zagreb, Croatia). We also wish to thank the participants in the reliability exercise and Wolfgang Grodd MD, Prof. of Neuroradiology, as well as Peter-Michael Weber, DTP laboratory, University Hospital Tübingen, Germany, for their support in the initial development of the neuroimaging chapter of the SCPE Reference and Training Manual.

The authors have stated that they had no interests which might be perceived as posing a conflict or bias.