Pediatr Radiol (2008) 38 (Suppl 3):S452–S458 DOI 10.1007/s00247-008-0846-5

REVIEW

Imaging of abdominal tumours: CT or MRI? Øystein E. Olsen

# Springer-Verlag 2008

Introduction

Choice based on empirical evidence

Is it at all possible to answer the question: computed tomography (CT) or magnetic resonance imaging (MRI) for abdominal childhood tumours? Those seeking a ‘yes’ or ‘no’ will be disappointed; the project is complex, and there are very few scientific stepping stones. The starting point is as always the more fundamental questions:

Empirical evidence is thought to be our ideal framework (‘evidence-base’). Defining efficacy as value, usefulness, potential, virtue, we want to find the exact efficacy of our investigations. It is undoubtedly possible that an image can have a virtue in being pretty and novel, it can be useful in its clarity of visualising something unexpected, and surely it may have direct implications for therapeutic decisions. Even with such a multifaceted view we can recognise that one achievement depends on other, e.g., that we cannot achieve diagnostic accuracy unless we can produce images that have an adequate signal-to-noise ratio (SNR). It is therefore possible to conceptualise all the different levels into a tiered model like this, based on a model by Fryback and Thornbury [1]:

1) What is the purpose of imaging? 2) Should we image, and when? 3) How do we acquire the optimal amount of information at lowest possible risk and cost? Clearly, our perception of what is achievable both in terms of potential benefit, and in terms of risk reduction is going to change year on year. The question is therefore difficult to conceptualise because it requires considerable judgement under constantly changing circumstances. Taking a few steps back we shall try and get a view of the principle factors that ideally should guide our practice, as well as the factors that probably are the essential determinants in the real world, hopefully ending up with a pragmatic compromise.

Dr. Olsen has disclosed that there are no conflicts of interest. Ø. E. Olsen (*) Radiology Department, Great Ormond Street Hospital for Children NHS Trust, Great Ormond Street, London WC1N 3JH, UK e-mail: [email protected]

1. An imaging technique must be feasible, and reliably produce images. 2. The technique must have a minimum technical efficacy, so that it delivers images with a certain signal-to-noise, contrast, sharpness, and little artefact, and we must be able to recognise the structures or (patho-) physiology of interest. 3. The investigation (interaction of images and readers) must be efficacious in terms of having a good sensitivity for detecting lesions and specificity for recognising normality. Slightly more advanced, we will require sufficient predictive values. 4. The procedure should have diagnostic thinking efficacy, meaning that the overall diagnostic process is influenced by the results of our investigations. 5. Therapeutic impact efficacy: our investigation impacts therapeutic choices. Therapeutic strategies may be changed as a result of the investigations; or radiology

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

may provide important justification for a continued therapy, which is naturally difficult to quantify. 6. Positive impact on complications of treatment, morbidity from disease, and mortality. 7. Cost-effectiveness in a societal perspective. It is clear that this model is hierarchical with item 1 as a foundation. It is hard to imagine any procedure being beneficial for public health unless it improves the outcome for individuals. It can only help our patients survive through influencing clinical management, and this in turn cannot be achieved unless the application of the test is accurate. Interestingly, a substantial foundation is required before we even get close to assessing the potential consequences of our test for the individual patient.

Risk assessment Having progressed through the tiered model and collected data for each step, we also need to add some important safety considerations, with the most important known risk factors being: 1) Delivery/absorption of ionising radiation (CT, nuclear medicine). Based on cohort studies, mostly in atomic bomb survivors, it is estimated that an excess of 50 deaths from cancer occur per one million individuals exposed to 1 mSv medical radiation, that is to say, one death per 20 Sv exposure [2]. Although the methodology of these studies, in particular their translational value, can be questioned, we have no alternative ways of estimating the risk. If true, a multidetector CT scan of the whole abdomen delivering in the region of 15 mSv translates into 7.2 years of background radiation, or an attributed risk of one death per 1,250 examinations. These estimates are probably low in a paediatric context, since it is thought that the lifetime risk equivalent is about three times as high in male and six times in female infants due to longer life expectancy and more radiation sensitive tissues [3]. 2) Thermal energy deposition (MRI). The energy delivered in the radiofrequency spectrum is known to induce temperature increases in tissues. Although the risk from this is not yet quantified, and it is unproblematic to operate below the recommended limits, prudence is essential as with any relatively young medical technology. 3) Intravenous contrast media (CT). The use of contrast media is essential since the inherent tissue contrast is very poor. With the development of faster scan technology, the precision of the bolus delivery is more crucial, which may call for higher doses of iodinated contrast medium (potential nephrotoxicity) as well as higher injection rates (risk of extravasation and line damage).

S453

4) Intravenous contrast media (MRI). Gadolinium chelates used as intravenous contrast in MRI is a triggering factor in the pathogenesis of nephrogenic systemic fibrosis in patients with reduced glomerular filtration rate (GFR). Several new governance principles have now been suggested, including an almost absolute contraindication when GFR<30, a relative contraindication when GFR is in the range 30–60, and preference of cyclical compounds [4]. Since, according to local experience, it is very difficult to achieve a diagnostic result without using gadolinium in abdominal imaging, this certainly means extra risk assessment obligations in paediatric MRI. It is essential to identify patients who may be at risk of having reduced kidney function (chronic kidney disease, recent nephrotoxic chemotherapy), or physiologically low GFR (infants), and in this group either test GFR specifically, or alternatively estimate GFR based on serum creatinine and height. 5) Risk associated with sedation and general anaesthetic (mostly MRI). Long scan times obviously make sedation/anaesthetic essential in paediatric MRI. Sedation should only be practiced under very strict regimens including selection of patients, induction, surveillance, and recovery; the aim is to bring the risk to a level where it is comparable to the very low risk of general anaesthesia [5]. In the ideal world having quantities for the efficacy at least to tier 4 in our hierarchical model, and being able to quantify the risk associated with CT and MRI, we would theoretically achieve: 1) Quantities for how far it would be possible to reduce imaging-associated risk factors without detrimental effect on our efficacy; e.g., how much can we reduce the radiation dose without risking image quality that is insufficient for us to maintain the desired efficacy at a higher tier? 2) A simple answer to the question of which modality is preferential. This is obviously over optimistic and over simplistic, as seen in the following evidence.

White spots on the empirical map Unless isolated in academia, the hierarchical efficacy model seems all too positivistic, if not utopistic. First of all, think of scientific literature in the light of the model. Most of us would intuitively respond that almost all existing data is from feasibility and technical studies (demonstrating the ability of visualising a pathological finding with a technique; difference in conspicuity of lesions between and old

S454

and new application; interobserver variability in recognising abnormality etc), and from accuracy studies (sensitivity and specificity compared to some ‘gold standard’; predictive value of findings). Unsurprisingly, a structured search of the two major knowledge databases (Excerpta Medica Database and Medical Literature Analysis and Retrieval System Online) yielded thousands of MRI studies published over the last 10 years; however, of the 66 papers on non-central nervous system paediatric oncological MRI, less than a third utilised any statistical analysis for objective assessment of results and even fewer studies tried to assess outcome beyond diagnostic impact (K. Park and Ø.E. Olsen, preliminary unpublished data). It is exceedingly rare in paediatric oncological imaging—presumably also in other sub specialties—to find published data from efficacy studies that reach beyond the third of the seven tiers of the efficacy model.

CT or MRI: real-world decision-making Having accepted that the ideal world description is a great concept, alas with fundamental breeches in its body of evidence, probably also with limitations even in the potential to achieve the desired knowledge, let us now explore how choices in all likelihood are made in the real world. Availability of technology Consider a few examples first: 1) Many centres now do lung CT in patients with Wilms tumour; however, the treatment protocols are still based on nodule visibility on chest radiographs, and the benefit of treating sub-centimeter nodules is highly speculative. 2) 64-channel CT scanners have started taking over from previous models. Image quality seems to improve, but is there actually any evidence in terms of outcome for the patient? 3) Quantitative techniques have become part of many standard scan protocols in MRI; however, what do the numbers really tell us? So do we perform examinations simply because we are able to? Although there may well be some reason behind applying these techniques, and surely we as radiologists have the patients’ best interests in mind, there are probably many further examples of modalities and imaging protocols that have not been scrutinised from an efficacy perspective. It is certainly too nihilistic to claim that we do something just because we can, and only to suggest this might be the case, sounds like professional suicide. However, if there is a

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

touch of truth in this, then the ultimate consequence is that we, to a large extent, are letting the equipment manufacturers lead the run. It may not be a bad thing, of course; many advances have materialised thanks to the availability of new techniques. The worry is that we have difficulty in quantifying the improvement, and also that clinical need is not necessarily the main driving force. Public expectations Here we experience opposing currents. On the one hand, there may be an expectation from political bodies that we justify our salaries in terms of ‘production’, which is usually seen with the eyes of an accountant, i.e., the number of procedures done; not the quality, which is too difficult to measure. Even patients or guardians may be expecting certain diagnostic procedures. On the other hand, we are increasingly facing scepticism, which may be due to media focus on the potential harm of ionising radiation. It is problematic to judge any of these currents of influence as inappropriate, but we are still lacking the evidence we need in order to get the balance right. Infrastructure The state of the current infrastructure reflects the interface between the strategy (and financial ability) of public or other funding bodies on one side, and radiologists’ local strategic choices on the other. Since nobody can ignore economic realities, priorities have to be set. Shall we establish a very expensive paediatric multidisciplinary MRI service, or is CT with a low-dose technique actually adequate? Who really knows? National and international guidelines Although recent years have seen more active participation from radiologists through societies and task forces, traditionally, protocols for diagnostic work-up and surveillance have been a small part of the treatment protocols in oncology, and, one may suspect, often devised by nonradiologists. Personal preference—radiologist and clinician Let us look at the term comfort zone. Nobody likes to step outside their own. The radiologist’s comfort zone may be defined by his/her familiarity with different modalities, both technical knowledge in terms of optimising the image acquisition, and in terms of image interpretation. Maybe we feel more comfortable with higher dose levels in CT, fearing to miss small lesions due to noisy images? Maybe our MR images have an uncomfortably low quality? Other

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

S455

staff groups must be comfortable, of course, and may not be entirely happy to handle sedated infants in the MRI scanner, or may not be adequately trained to perform high-quality diagnostic US scans. How about clinical colleagues’ comfort zone, for example the surgeon who ‘needs’ a CT rather than a perfectly adequate MRI before taking the patient to theatre for major tumour surgery; or any other clinician who has the ultimate responsibility for patient management? Balancing the conservatism of the comfort concept, are professional interests and drives, the pioneering mind, and research interest. All these factors are surely not irrelevant, although possibly not completely appropriate.

honoured with many achievements in the development of better care for our patients. On the downside, however, a strong influence of individuals and commercial interests makes it much more difficult to set out a clear strategic direction. There is also a danger that practice is rendered too heterogeneous (or personal, idiosyncratic). As the discussion so far has demonstrated, we may wish to base our practice on empirical evidence, but we are forced to accept that deductive thinking, not to say “art”, will remain an important characteristic of our field. In a pragmatic approach we may attempt to merge the strict empirical ideal with real world factors. In an attempt to compromise we may end up with a list of decision-factors ordered by decreasing desired importance:

Towards a pragmatic compromise

1) Empirical evidence of outcome efficacy from structured assessments. 2) Safety (including potential risks from ionising radiation). 3) Compliance with peer-agreed best-practice guidelines, or with carefully deduced evidence of efficacy. 4) Existing local radiological infrastructure, expertise and funding. 5) Existing local institutional infrastructure, expertise and funding. 6) Personal preference and public expectations.

We have so far witnessed a polarised project: the ideal frameworks of empirical evidence and clinical governance on one side; the reality in an imaging department, partly in response to the lack of empirical evidence, on the other. However, this dichotomy is not necessarily valid, and there may be strengths and weaknesses with both approaches. Precise empirical facts would naturally solve most problems and surely answer our introductory questions. It would provide a framework for understanding exactly how to investigate and would help us balance cost (financial, radiation risk, other risks, patient discomfort) against known benefits (efficacy). The main problem is, as seen, that the facts are not there. A major criticism may also be that we may never achieve complete knowledge, that is, unless the technological development ceases. Should we hypothetically attain perfect knowledge at some point, this could only be through structured and controlled investigations, which poses another uncomfortable question: would we at all be able to replicate the controlled research environment in a clinical department? We should acknowledge that from the accuracy level and up, the yield of a technique is highly dependent on the user (operator, reader etc). To spell it out: a radiologist who follows the child through the diagnostic process, who excels in sonography, may in practice perform an US study that is much more efficacious than a CT or MRI replicating some controlled research conditions. The real world, although seemingly bleak from the altruistic empiricist's viewpoint, may still hold some positive qualities. It first of all seems to accommodate the fact that technological and clinical potentials are changing so rapidly that it might sometimes be unethical to sit back and wait for scientific proof before translating technology to clinical practice. The interaction between technological development and rapid clinical implementation is a strong driving force (for whatever reason) and can probably be

At a very self-critical moment, one may insinuate that this list describes exactly how our practice is currently shaped, but in inverted order! That is, apart from safety considerations, which hopefully are always at the front end of any decision process. There may be some truth in this; however, I am not suggesting any kind of professional incompetence or malpractice. If only partly true, it is an incentive to implementing change. There are clearly at least two strategies to this. Most obviously, we should encourage empirical structured assessments. A complementary strategy seems much more feasible, namely to systematically identify ways of minimising the effects of perceived lowpriority factors (items 4–6 in the list), most of which are local. Some action points may relate to the concept of comfort zones as discussed above: 1) Radiologists’ more direct involvement in image acquisition, for example in body-MRI, might lead to a better understanding of, and more reliance on, this very operator-dependent modality. 2) Improved communication between clinician and radiologist of findings from heavily operator-dependent modalities (ultrasound (US) and MRI) may relieve some of extra-radiological discomfort with US and MRI. Other factors relate to local infrastructure. 3) Setting up an efficient sedation and anaesthetic service may reduce waiting times for MRI and channel more

S456

patients away from CT. Another important factor is cross-institutional collaboration. 4) Defining acceptable ranges for radiation exposure. 5) Promoting low-radiation techniques or non-ionising radiation alternative imaging.

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

both the diagnostic and therapeutic thinking via a set of fundamental capabilities:

Attempting to deduce some evidence of efficacy of CT and MRI in abdominal tumour imaging, we need to consider the body of scientific knowledge and personal experience regarding CT and MRI, and carefully extrapolate the facts. At this point we assume that adequate imaging benefits

1) Ability of high spatial resolution. This is inherent in CT imaging. In MRI, the spatial resolution is decided by the operator, and can theoretically be as high as desired, but unfortunately there is an inverse relationship between spatial resolution and signal-to-noise ratio. This is due to the fact that the number of spins contributing to each pixel depends on the chosen voxel size, whereas the noise in the system is constant for any set coil configuration. For a given pulse sequence, the loss in SNR may be compensated by increasing the acquisition time, however, only partially because SNR is proportional to the square root of acquisition time. It requires careful adjustment to be able to replicate the excellent spatial resolution of CT with MRI, but it is mostly possible in the abdomen (Fig. 1).

Fig. 1 Comparison of spatial resolution, lesion conspicuity and vascular relations between CT (left column) and MRI (right column) in a young child with a retroperitoneal neuroblastoma (arrowheads). Intravenous contrast medium is used in both modalities to visualise

the relatively poorly enhancing tumour against the enhancing vessels and surrounding anatomy. Selected images at the level of the coeliac axis (top row), superior mesenteric artery (middle row) and renal arteries (bottom row). Lk left kidney

These are only some examples based on own experience, and perhaps rather self-evident, but the list can easily be extended.

Evidence deduced

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

Fig. 2 MRI in a young child with a left suprarenal neuroblastoma and retroperitoneal infiltration across the midline. a Coronal STIR spinecho image (TI/TR/TE 130/4,000/15 ms) performed with respiratory triggering demonstrates the extent of the lesion (arrowheads). b This transverse image shows enhancement (red overlay from a 3D-FLASH [TR/TE/FA 4.4/2.2 ms/25°] arterial-phase contrast-enhanced acquisition) on the background of a STIR image (TI/TR/TE 130/4,500/ 150 ms). c This image shows apparent diffusion coefficients (mapped

2) Ability of useful image contrast. The inherent contrast in CT is poor and only adequate at tissue–gas interfaces and tissue–bone/calcium interfaces, unless we use iodinated contrast agents. The number of different image contrast (phases) acquired is limited due to the radiation dose penalty, and it is generally agreed that only one acquisition is justified. MRI on the other hand can produce a multitude of tissue contrasts that can even be co-registered (Fig. 2), and a number of phaseseparated acquisitions following contrast medium delivery (Fig. 3). 3) Ability to produce images with minimal artefact. With the last generations of multichannel CT equipment, the burden of motion artefact is substantially reduced, and often negligible. Most MRI artefacts are easily dealt with; however, motion artefact can be difficult to minimise because shorter scan times mean lower SNR as seen above. Finding the ideal pulse sequence and scan time, and utilising the motion compensation technologies of modern scanners is possible, but again requires a skilled operator. 4) Ability to produce useful quantitative data. At the moment, CT is limited to volume estimates/measurements. This may seem adequate from the point of view of current treatment protocols, but size may not be a universal reliable response criterion in paediatric

S457

as a red overlay, based on an echo-planar spin-echo diffusion-weighted sequence with b-values of 0, 50, 250, 500, 1,000) on a background of a STIR image (TI/TR/TE 130/4,500/15 ms) highlighting areas of restricted diffusion (bright orange). Such co-registration is very easily performed both on commercial workstations and with free software (e.g., Osirix; www.osirix-viewer.com), and may facilitate the interpretation of a large amount of data

Fig. 3 MRI in a young child with hepatoblastoma. Both images (5-mm thin MIP) were acquired with 3D-FLASH (TR/TE/FA 4.4/2.2 ms/25°) during breath-hold and demonstrate the lesion (arrowheads). a Acquisition in arterial/portal-venous phase depicts the coeliac axis and two hepatic arteries (arrows). b Additional acquisition in venous phase demonstrates the relation of the tumour to the middle and right hepatic arteries (long arrows) and the main portal branching (short arrow)

S458

tumours [6]. MRI can produce quantities for perfusion, T1, T2 and apparent diffusion coefficients. The latter is known to correspond to the cellularity of a lesion [7], however it remains to be seen whether it may be a useful response criterion. 5) Ability to cover a large anatomical area. Both modalities would be adequate, but CT obviously carries a radiation dose penalty. 6) Feasibility of repeat imaging. CT has a higher nonoperator dependent reproducibility; however, the accumulated radiation dose from diagnostic, followup and surveillance scans cannot be ignored. MRI is feasible; however, it requires operator reliance, and potentially sedation or general anaesthetic each time. During nephrotoxic chemotherapy there may be limitations in the use of gadolinium chelates due to reduced GFR.

Conclusion Appropriate imaging requires a plethora of empirical evidence, which is still lacking. In current practice therefore one may suspect that personal preference and expertise, local infrastructure, and a political/managerial expectation to “production” play inappropriately important roles in the decision-making process, including which modality to choose in oncological care. Whereas many of these factors may be justified, e.g., due to very scarce human and financial resources, the deduced evidence of efficacy as well as risk profiles would favour a stronger emphasis on

Pediatr Radiol (2008) 38 (Suppl 3):S452–S458

MRI in abdominal imaging given that some very important prerequisites are adequately addressed: 1) Reliable operation of MRI, which ultimately rests on the radiologist, 2) An efficient and safe service for sedation and anaesthesia, 3) Appropriate governance to reduce the risk of gadoliniumassociated complications, and 4) Effective presentation and communication of imaging interpretation to clinicians. It is still the case that a good CT is better than a substandard MRI, but there is hardly any excuse for accepting poor performance.

References 1. Fryback DG, Thornbury JR (1991) The efficacy of diagnostic imaging. Med Decis Mak 11:88–94 2. International Commission on Radiological Protection (1991) ICRP publication 60: 1990 recommendations of the international commission on radiological protection, 60. Elsevier, Amsterdam 3. Mayo JR, Aldrich J, Muller NL et al (2003) Radiation exposure at chest CT: a statement of the Fleischner Society. Radiology 228:15–21 4. Mendichovszky IA, Marks SD, Simcock CM et al (2007) Gadolinium and nephrogenic systemic fibrosis: time to tighten practice. Pediatr Radiol (in press) 5. Sury MR, Smith JH (2008) Deep sedation and minimal anesthesia. Paediatr Anaesth 18:18–24 6. Olsen OE, Jeanes AC, Sebire NJ et al (2004) Changes in computed tomography features following preoperative chemotherapy for nephroblastoma: relation to histopathological classification. Eur Radiol 14:990–994 7. Humphries PD, Sebire NJ, Siegel MJ et al (2007) Tumors in pediatric patients at diffusion-weighted MR imaging: apparent diffusion coefficient and tumor cellularity. Radiology 245:848–854