Article Text

Download PDFPDF

The value of magnetic resonance spectroscopy in tumour imaging
  1. Andrew C Peet1,2,
  2. Theodoros N Arvanitis2,3,
  3. Dorothee P Auer4,
  4. Nigel P Davies1,5,
  5. Darren Hargrave6,
  6. Franklyn A Howe7,
  7. Tim Jaspan8,
  8. Martin O Leach9,
  9. Donald Macarthur8,
  10. Lesley MacPherson2,
  11. Paul S Morgan4,
  12. Kal Natarajan1,5,
  13. Geoffrey S Payne9,
  14. Dawn Saunders10,
  15. Richard G Grundy11
  1. 1
    Academic Paediatrics and Child Health, University of Birmingham, UK
  2. 2
    Birmingham Children’s Hospital Foundation Trust, UK
  3. 3
    Electrical, Electronic and Computer Engineering, University of Birmingham, UK
  4. 4
    Academic Radiology, Nottingham University Hospitals, UK
  5. 5
    Medical Physics and Imaging, University Hospital Birmingham Foundation Trust, UK
  6. 6
    Royal Marsden Hospital Foundation Trust, UK
  7. 7
    St George’s, University of London, UK
  8. 8
    Nottingham University Hospitals, Nottingham, UK
  9. 9
    Institute of Cancer Research, Sutton, UK
  10. 10
    Great Ormond Street Hospital for Sick Children, London, UK
  11. 11
    Children’s Brain Tumour Research Centre, Nottingham University Hospitals, Nottingham, UK
  1. Andrew Peet, Chair CCLG Functional Imaging Group, Academic Paediatrics and Child Health, Whittall Street, Birmingham B4 6NH, UK; acpeet{at}

Statistics from

Magnetic resonance (MR) imaging has a key role in the management of many childhood tumours. There is increasing interest in extending these investigations to MR techniques that give information on tumour biology in vivo. Magnetic resonance spectroscopy (MRS) is one such method and it provides information on tissue biochemistry. Promising results have been obtained from many preclinical and clinical studies, leading to an expectation that MRS will play a valuable clinical role. However, the role of MRS is not yet well defined and there is a paucity of data from multi-centre clinical trials. In this article we concentrate on MRS in paediatric oncology and provide some general guidance on current applications and outline areas that need to be developed further.


Certain atomic nuclei (eg, 1H, 31P and 13C) possess a magnetic moment and when placed in a strong magnetic field will resonate at a particular radiofrequency that subtly depends upon the chemical environment. In MR spectroscopy, the frequencies and intensities of these resonances are measured and represented graphically in an MR spectrum. The most commonly available method clinically is 1H MRS, and the 1H spectrum is a biochemical profile of the small mobile metabolites and macromolecules present in the tissue. 1H MRS can be performed with a standard clinical MRI scanner as part of a conventional MRI investigation. An example 1H spectrum from normal brain is given in figure 1. The horizontal scale, in units of parts per million (ppm), represents signal frequency adjusted to be invariant to the strength of the magnetic field of the MR scanner. Each metabolite is identified by one or more peaks at specific ppm values with the areas under the peaks being proportional to the metabolite concentration. The main metabolites observed are N-acetyl aspartate (NAA), creatines (Cr), glutamate/glutamine (Glx), myo-inositol (mI), cholines (Cho), macromolecules (MM) and in disease, lactate (Lac) and mobile fatty acids (Lip). The technical issues of detecting metabolites at millimolar concentrations in the presence of tissue water at 41 molar has meant that currently 1H MRS studies are predominantly of the brain, although studies of the breast and prostate in adults are becoming more common. The region of tissue from which 1H MRS data is acquired is selected using conventional MR images. In single voxel spectroscopy (SVS), a single volume of tissue, termed a voxel, is selected. A voxel of 4 to 8 cm3 will provide a spectrum in just 5 mins of data acquisition. Smaller voxels require increasingly prohibitive acquisition times. Multivoxel spectroscopy (MRSI) acquires data simultaneously over a two-dimensional or three-dimensional grid of voxels and is more efficient but can take from 10 to 30 min, is technically more difficult and is less reproducible. The appearance of 1H MR spectra is highly dependent on the precise method of data acquisition, specified by a set of acquisition parameters. The most important parameter is the echo time. Short echo times provide information on more metabolites. 31P MRS requires additional scanner hardware and being less sensitive than 1H MRS requires larger tissue volumes of interest. 31P MRS allows adenosine triphosphates (ATP), phosphocreatine (PCr), inorganic phosphate (Pi), phosphomonoesters (PME) and phosphodiesters (PDE) to be detected and intracellular pH calculated.

Figure 1 A short echo time (30 ms) 1H MRS from normal brain acquired using single voxel spectroscopy on a 1.5 Tesla scanner. The main metabolites observed are N-acetyl aspartate (NAA), creatine (Cr), glutamate/glutamine (Glx), myo-inositol (mI), cholines (Cho), macromolecules and fatty acids (MM+Lip).


There have been numerous studies in adults with brain tumours. 1H MRS with pattern recognition has shown diagnostic accuracy of over 92%1 2 and a 15% improvement in the number of correct diagnoses compared to MRI alone.3 Murphy et al found that 1H MRS made a significant contribution to the pre-operative diagnosis in 6%4 of patients. Large multi-centre projects have been set up to develop computer software for aiding 1H MRS-based tumour diagnosis and include INTERPRET,5 eTumour6 and HealthAgents.7 The eTumour and HealthAgents projects explicitly include paediatric data. Fewer studies have specifically concentrated on children with brain tumours. Hourani et al showed in a study of 32 children that a high Cho/Cr can distinguish between brain tumours and other CNS lesions with 78% accuracy.8 A more common clinical problem is making a non-invasive diagnosis of the histological sub-type of a tumour. Wang et al measured NAA/Cho and Cr/Cho using 1H MRS and combined these with a pattern-recognition technique to classify the major cerebellar tumour types in 31 children, achieving an accuracy of 85%.9 Panigrahy et al used 1H MRS to investigate 60 children with brain tumours.10 They found significant differences in several metabolites for the major histological sub-types and proposed that this could be a useful non-invasive aid to diagnosis. These studies used specialist data-processing techniques. More recently, Harris et al developed a method for analysing 1H MRS of cerebellar tumours that could potentially be applied directly in a clinical setting.11 Overall, there is sufficient data in the literature to interpret the main 1H MRS features of common childhood brain tumours to aid diagnosis. Additional metabolites may be measured using 31P MRS, and Albers et al demonstrated the value of this for cerebellar tumours.12 Combining improved metabolite profiling with sophisticated analysis software will undoubtedly improve the diagnostic accuracy of MRS for children with brain tumours in the future. However, it will be important to determine the predictive accuracy of these non-invasive diagnostic tools in large multi-centre prospective trials and demonstrate their use in a clinical environment.


Prognostic markers are used to stratify patients into different treatment groups, and the development of non-invasive prognostic biomarkers would be of particular value. Many studies have investigated Cho as a marker of poor prognosis; Girard et al found that Cho/NAA was a marker of poor prognosis in childhood cerebral hemisphere tumours at diagnosis,13 and Warren et al established that this was also true for recurrent tumours.14 Astrakas investigated 66 children with brain tumours using 1H MRSI and found that both Cho and lipids were useful for grading a range of childhood brain tumours.15 This cohort of patients has been followed up to show that this biomarker also predicts poor prognosis across a range of different tumours.16 The recent appreciation of lipid levels as a marker of tumour malignancy is consistent with grading of adult gliomas.17 18 To date, the studies have included patients with a variety of diagnoses treated with a number of different treatment regimes. Before this technique can be used in a clinical setting we need to evaluate specific patient subgroups treated in the same way, ideally within the context of a clinical trial. A single centre study of medulloblastoma detected key differences between metastatic and non-metastatic cases in the Cho/lipid ratio.19 However, preliminary data indicating that metabolic profiles characterise tumour biology and can predict tumour behaviour need to be confirmed by multi-centre trials where patients are selected and treated in a uniform manner.


In addition to providing biomarkers of prognosis, MRS has been investigated as a method for treatment monitoring with the aim of developing a marker of early response or relapse. Lazarref et al used 1H MRSI serially during and after treatment in 10 children with gliomas and found that an increase in tumour Cho was a good marker of progression.20 Conventional MRI is a poor indicator of tumour status for diffuse pontine gliomas treated with radiotherapy,21 and therefore a potential clinical area for MRS is to improve monitoring. 1H MRSI has been used to investigate diffuse pontine gliomas in eight children at diagnosis, after radiotherapy and at relapse.22 Cho/NAA, lactate and lipids decreased after radiotherapy and increased again at relapse; importantly three patients showed MRS changes prior to clinical and radiological deterioration. A recent study of 16 children has confirmed the ability of 1H MRS to predict tumour progression of diffuse pontine gliomas prior to enlargement of the tumour on conventional MRI.23 These findings now need to be confirmed in a multi-centre study and the place of 1H MRS in the management of diffuse pontine gliomas more clearly defined.


A technique that has shown particular promise clinically in adults with brain tumours is MRSI-aided biopsy.24 25 Most of the experience is in high-grade gliomas where heterogeneity and necrosis within the tumour can lead to non-diagnostic biopsies. The technique relies on the observation that Cho is raised while NAA and Cr are reduced in tumours compared with normal brain. Maps of Cho/NAA and Cho/Cr identify regions of active tumour. This method has not been studied thoroughly in children, but MRS could be useful in delineating active areas of tumour in large heterogeneous lesions, particularly if combined with other MR techniques such as perfusion imaging. The optimal metabolite ratio and most appropriate cut-off values need to be considered carefully as some childhood tumours do not have a consistently high Cho/Cr ratio.10 11


Treatment monitoring by 31P MRS has also been studied in children and adults and changes detected in some of the major metabolites particularly PME and PDE. Studies have concentrated on sarcomas of the extremities26 27 and lymphoma28 although a seminal study was carried out on two patients with neuroblastoma.29 31P MRS would need to be more widely available to enable robust evaluation in clinical practice.


While MRS is a powerful tool in its own right, it may be readily combined with other MR techniques such as diffusion weighted imaging (DWI), diffusion tensor imaging (DTI), dynamic contrast enhanced MRI (perfusion imaging) and functional MR in addition to conventional structural MR imaging. A recent study has combined 1H MRS with DWI to provide accurate classification of posterior fossa tumours, albeit with small numbers in each tumour class.30 1H MRS has also been combined with perfusion imaging to provide enhanced indicators of tumour progression31 while multiparametric analysis has also been explored.32 Determining the optimal multi-modality imaging paradigm will be a major goal in the future, which may enable more accurate pre-operative tumour assessment, assist in surgical planning/strategy and provide more effective evaluation of therapeutic interventions.


The metabolic information provided by MRS adds value to the differential diagnosis, prognosis and treatment evaluation for certain childhood cancers. However, its use is still largely restricted to centres with technical expertise, particularly in data analysis. New anti-cancer drugs are commonly aimed at specific molecular targets and the potential of MRS to provide a direct measure of drug activity is becoming increasingly pertinent.33 While the areas where MRS may aid clinical practice are becoming better defined, it is unlikely that these benefits will become widely accepted or available until they have been evaluated in robust multi-centre trials. A multi-model approach combining metabolic information with diffusion and perfusion characteristics is likely to yield the most benefit for improved understanding and management of childhood cancer.


The Functional Imaging Group’s activities have received funding from Cancer Research UK, The European Union Framework 6 project HealthAgents (IST-2004-27214), the Samantha Dickson Brain Tumour Research Trust, Birmingham Children’s Hospital Research Foundation and the Joe Foote and Alistair Wainwright Trusts. The references cited in this article were selected by consensus within the authorship.



  • Funding: All authors are active members of the Children’s Cancer and Leukaemia Group’s Functional Imaging Group.

  • Competing interests: The Functional Imaging Group’s meetings have received part funding from MR scanner manufacturers.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.