Introduction

The delivery of personalized medicine demands the availability of robust and reliable biomarkers. Although these may be genetic, serum or functional imaging parameters, only the last of these have the capacity to provide information on change through the course of the disease and on disease heterogeneity both among and within lesions. Thus, imaging biomarkers are invaluable in providing information not just for diagnostic purposes at the outset, but also on informing clinical decision-making at various points in a treatment pathway.

The availability of functional imaging biomarkers has superseded the traditional concept of imaging as a purely morphological tool that provides information on the size or volume change of a tumour with time and treatment. Aspects of a tumour such as its changing vascular profile, water content, degree of apoptosis or necrosis or metabolism are all now measurable through advanced imaging techniques using magnetic resonance imaging (MRI), computed tomography (CT) and radioisotope studies with SPECT and PET. Techniques such as dynamic contrast-enhanced (DCE) imaging allow derivation of semi-quantitative parameters (such as the enhancement fraction), but pharmacokinetic modelling allows estimation of quantitative parameters such as permeability and wash-out. Likewise, diffusion-weighted MRI (DW MRI) provides information about the cellularity of tumours, as well as the change in the extracellular water compartment in the face of increasing apoptosis or necrosis. MR spectroscopy interrogates metabolite levels within tumours and their relative changes with tumour evolution and treatment. This review focuses on the functional techniques currently available in the clinic with MRI and CT, and explores the difficulties and limitations in measurement that need to be addressed for successful implementation of the imaging biomarker in a clinical setting.

The tissue organization, vascular and metabolic biomarkers discussed in this article are summarized in Table 1.

Table 1 Tissue organization, vascular and metabolic biomarkers discussed in this article
Full size table

Tissue organization biomarkers

Apparent diffusion coefficient

Mechanisms of contrast

DW MRI exploits the incoherent motion of water molecules within tissues to generate contrast. Many solid tumours exhibit restricted diffusion of water molecules compared to many normal tissues, leading to bright signal on diffusion-weighted images and low values of apparent diffusion coefficient (ADC) [1, 2].

Following the initial radiofrequency (RF) excitation, two or more diffusion-encoding gradients are applied along a specified direction, separated by a refocusing RF pulse. The diffusion-encoding gradients induce a loss of phase coherence, and hence loss of signal, of the protons which have moved in the direction of the gradient during the diffusion encoding time (typically a few tens of milliseconds). The diffusion-weighted images show lower signals from voxels where the water molecules diffuse freely and higher signals from voxels where diffusion of water molecules is more restricted.

The strength of the diffusion-weighting is determined by the magnitudes and timings of the diffusion-weighting gradients and is commonly described by a summary parameter known as the b-value [3]. Estimates of ADC may be derived from fitting a monoexponential function to the signal (S) measured at two or more b-values (Eq. 1).

$$ S(b)={S}_0 \exp \left(-b.ADC\right) $$
(1)

Restricted diffusion within tumours has been shown to be related to increased cellularity and reduction in extracellular space [1, 2, 4]. Increases in necrosis and cell death after treatment increases ADC [1, 2], which has been shown to be predictive of response to chemotherapy in various tumour sites [4, 5].

Technical requirements

DW MRI can be carried out on most modern MR scanners and does not require administration of exogenous contrast agents. Echoplanar imaging (EPI) is usually employed to reduce sensitivity to motion [3]. EPI is, however, sensitive to field inhomogeneities and chemical shift artefacts and optimization of sequence parameters is required to obtain good quality images [6]. Good B0 homogeneity, minimal eddy current effects, high signal-to-noise ratio (SNR) and good fat suppression are required [6]. ADC maps are provided by the manufacturers’ software. Many studies also use in-house software for definition of regions of interest (ROIs) and calculation of ADCs.

Validation, qualification and clinical applications

Repeatability studies have shown within-patient coefficients of variation (wCV) between 4 % and 15 % [7–9]. Multicentre studies of healthy volunteers have shown significant differences in ADC estimates between scanners from different manufacturers in some abdominal organs [10] and in grey matter and white matter [11].

ADC has been shown to be negatively correlated with histological measures of cell density in glioma [12] and in colorectal liver metastases [7]. A negative correlation has also been shown between ADC and the proportion of collagenous fibres in pancreatic cancer [13]. A study of 32 patients with locally advanced gastrooesophageal cancers showed a correlation between the change in ADC estimates after neaoadjuvant treatment and the tumour regression grade determined from histology [14]. Although an increase in ADC estimates between pretreatment and posttreatment measurements has been shown to be predictive of response to treatment in many cases, some studies have not demonstrated correlation between change in ADC and response [5].

The use of DW MRI as a biomarker has revolutionized oncological diagnosis. As it can be used both qualitatively by viewing the high b-value images and ADC maps as well as quantitatively to generate mean or median tumour ADCs, it has been exploited as a diagnostic tool, and as a prognostic/predictive biomarker as well as for longitudinally monitoring treatment response. In several tumour types, e.g. liver [15, 16], lung [17], kidney [18], breast [19], prostate [20] and cervix [21], it is used to differentiate tumour from nontumour tissue (Fig. 1). However, its quantitative potential has been exploited to predict the aggressiveness of disease in prostate cancer [22] and histological grade in renal cancer [23] and cervix cancer [24]. Change in ADC has been shown to be predictive of therapeutic response in cervix cancer [25], colorectal liver metastases [26] and ovarian cancer [8] (Fig. 2) as well as for predicting local recurrence in rectal cancer [27], endometrial cancer [28] and biochemical recurrence in prostate cancer [29]. More recently, whole-body DW MRI has become possible through advances in RF and gradient coil technology as well as software for integrating image stacks into a visual representation of the whole body in a 3D multiplanar reformat. This type of image has been used for metastases screening, particularly in patients in whom bone lesions may be the only site of disease and bone scintigraphy is negative, e.g. multiple myeloma [30]. In this whole-body mode it has the advantage of being able to estimate the total tumour burden for skeletal metastases (Fig. 3) and follow their response to treatment, which has hitherto not been possible [31].

Fig. 1
figure 1

Prostate cancer in an 82-year-old man. a Coronal and b transverse T2-weighted images through the prostate apex with c corresponding transverse ADC map and d enhanced image 60 s after gadolinium administration show an anterior prostate tumour (arrows). The tumour is difficult to appreciate in a and b, but easily identified in c and d. The spatial mismatch between the regions of restricted diffusion in c and contrast enhancement in d is attributed to heterogeneity of the biological mechanisms determining these features within the tumour

Full size image
Fig. 2
figure 2

Ovarian cancer in a 70-year-old woman. a Transverse T2-weighted, b diffusion-weighted (b = 1,050 s/mm2) and c ADC map through the lower pelvis show irregular soft tissue with marked diffusion restriction on the right peritoneal reflection and serosal surface of the sigmoid colon (arrows). At a higher level in the mid pelvis d transverse T2-weighted, e diffusion-weighted (b = 1,050 s/mm2) and f ADC map show bilateral peritoneal deposits (arrows). These are best seen in e and are quantifiable from f, but are less conspicuous in d

Full size image
Fig. 3
figure 3

Whole-body diffusion-weighted image in a 67-year-old woman with biopsy-proven multiple myeloma. The coronal plane is reconstructed from data acquisition in the transverse plane in multiple stacked volumes and displayed as an inverted grey scale of the b = 1,050 s/mm2 images. Focal areas of restricted diffusion are noted throughout the skeleton indicative of focal sites of myelomatous deposits within bone marrow

Full size image

Current limitations and future perspectives

A major limitation in the validation and qualification of ADC as a biomarker in oncology is the lack of standardization for data acquisition and analysis. There are currently no standard sequences which can be implemented on all platforms which limits the extent to which data acquisition can be standardized in multicentre projects [6, 32]. Technical limitations, for example non-uniformity in ADC estimates, may introduce errors, particularly when employing large fields of view [32]. Software and methods for analysis are not standardized, leading to variation in definition of ROIs and calculation of ADCs [33].

The physiological basis of the diffusion-weighted signal is not fully understood and validation, for example by correlation with histopathology, remains an area of current and future work. The temporal evolution of ADC in response to treatment may be influenced by factors such as cell swelling, cell shrinkage, necrosis, fat infiltration and fibrosis [2, 4]. Moreover, attempts to use pretreatment ADC estimates as a predictive biomarker have yielded mixed results with some studies showing a correlation between pretreatment ADC and response to treatment while many other studies have shown no correlation [5]. A limitation of many studies is that numbers of patients have been small and meta-analyses have been impeded by differences between imaging protocols, patient populations and treatment regimens [5].

Current work is focused on qualification of ADC as a biomarker in oncology drug development [4, 34]. Work includes optimization and standardization of data acquisition, assessment of ADC as a prognostic or diagnostic biomarker in multicentre studies and histopathological validation [4, 6].

Magnetization transfer ratio

Mechanisms of contrast

Magnetization transfer (MT) contrast in MRI arises from interactions between protons (hydrogen nuclei) in a bound pool, for example protons attached to macromolecules or in hydration layers, and protons in a free pool, for example mobile water protons [35]. Bound pool protons do not contribute directly to the measured signal in most conventional MR imaging (which measures freely mobile protons) owing to their short T2 relaxation times (<1 ms) [36, 37]. The presence of protons in the bound pool may be detected via their interactions with the free pool by selectively saturating the broad resonance of the bound protons using an off-resonance RF pulse. This transfers saturation to the free pool by exchange of spins between the bound and free pools, either by chemical exchange or by dipolar coupling, and reduces the longitudinal magnetization available for imaging. This results in a reduction in signal intensity, compared to that observed without the off-resonance saturation pulse. Magnetization transfer ratio (MTR), which is the difference between the signal intensity observed without the off-resonance saturation pulse (M0) and the signal intensity observed after the saturation pulse (MSAT), normalized to M0, is used to quantify the MT effect (Eq. 2).

$$ MTR=\frac{M_0-{M}_{SAT}}{M_0} $$
(2)

Technical requirements

MT sequences are available on most modern MR scanners and do not require additional hardware or software. MTR maps may be produced using manufacturers’ software. As with any subtraction technique, MTR measurements are sensitive to motion and techniques such as breath-holding or cardiac gating may be required. Safety considerations, which require the specific absorption rate to be kept within regulatory limits, restrict the flip angle of the saturation pulse. Agar and albumin have been used to provide stable and reproducible MT phantoms for protocol development [37, 38].

Validation, qualification and clinical applications

Unlike DW MRI, there has been no attempt at standardization of MTR for use as a biomarker in oncology. However, validation against histopathology has been reported both in preclinical studies and in neurological applications in the clinic. MTR has been correlated with histological measures of fibrosis in pancreatic tumour models [39] and with collagen content in meningiomas [40] and with volume fraction of tumour cell nuclei in astrocytomas [41] in clinical studies. One multicentre study of the brains of healthy volunteers showed significant variation in MTR histogram metrics, which the authors suggested may have been due to the difference in flip angle of the off-resonance RF pulses between scanners [42].

Current limitations and future perspectives

MTR is still in the early stages of validation and qualification. MTR depends on pulse sequence and magnetic field strength [36, 37, 48] and the lack of standardization of MT sequences between scanners may contribute to differences in measurements of MT properties [42]. Development of MTR as a biomarker therefore also requires assessment of variability between scanners and standardization of sequences.

Although many studies have demonstrated the use of MTR in discriminating between benign and malignant lesions and among grades of tumours, further investigation of MTR is required in assessment of response to treatment. More advanced modelling may be used to extract other parameters to prove its utility as a response biomarker [35].

In clinical trials, two studies of brain tumours have shown higher MTR in high-grade astrocytomas than in low-grade astrocytomas [40] and in high-grade gliomas than in low-grade gliomas [41]. However, another study has shown that while tumours, infection and infarction all have significantly lower MTR than normal grey matter and white matter, the difference between MTR of high-grade and low-grade gliomas is not significant [43]. Clinical studies outside the brain include breast lesions where MTR has been used to discriminate between benign and malignant histologies [44], parotid glands [45] and prostate (Fig. 4). A feasibility study of MT imaging in patients with non-small-cell lung cancer has shown similar structures in MTR maps and FDG PET images and showed lower MTR in areas of suspected atelectasis than in tumours [46].

Fig. 4
figure 4

Prostate cancer in a 63-year-old man. a Transverse T2-weighted image through the mid-prostate shows a low signal intensity tumour in the right peripheral zone (arrow). b Corresponding T2 map, c ADC map (calculated from data acquired at b = 0 s/mm2 and 800 s/mm2) and d magnetization transfer contrast image (subtraction of data acquired without and with a 1,000 Hz off-resonance pulse) show a short T2 lesion (b, arrow) with restricted diffusion (c, arrow) and very little magnetization transfer effect (d, arrow) compared to the peripheral zone on the left

Full size image

MTR has also been used in the detection of postradiation fibrosis. A study of patients with rectal cancer treated with neoadjuvant chemoradiotherapy showed that the mean MTR of regions of fibrosis is significantly higher than the MTR of residual tumour, normal rectal wall or oedematous rectal wall; comparison with histopathological analysis showed that MTR diagnosed fibrosis with a sensitivity of 88 % and specificity of 90 % [47].

Native T1 relaxation time

Mechanisms of contrast

T1 relaxation time is a time constant for the recovery of longitudinal magnetization following excitation. T1 depends on the molecular environment of water molecules as well as on magnetic field strength [49]. Native T1 refers to the longitudinal relaxation time of protons in the absence of exogenous contrast agents. Changes in native T1 may be indicative of alterations in oedema [50] or release of paramagnetic ions and proteins during the destruction of cells and tissues [51].

Technical requirements

Inversion recovery and saturation recovery methods can be used to estimate T1 but acquisition times are relatively long [52]. Spoiled gradient-echo sequences with two or more flip angles are often used when faster acquisitions are required [53]. Optimization of flip angles is required for the range of T1 values to be measured [54]. Inhomogeneities in transmit B1 field can cause errors in T1 estimates values due to errors in flip angles [55].

Validation, qualification and clinical applications

Native T1 of tumours has been shown to correlate with decreases in tumour volume in a TH-MYCN transgenic mouse model of neuroblastoma treated with cyclophosphamide, a vascular disrupting agent (ZD6126) or an antiangiogenic agent (cediranib) [56]. Similar correlations have been observed in two mouse tumour models treated with an mTOR inhibitor (everolimus) [51] and in eight tumour models treated with five different chemotherapeutic agents where 15 – 20 % reductions in native T1 were seen in all cases of successful chemotherapy compared to controls, but not in drug-resistant models [57]. However, this needs to be considered in the context of reproducibility: a study of patients with metastatic colorectal cancer found a wCV of 15.8 % for baseline T1 measurements [58].

Only a small number of published studies have assessed native T1 as a biomarker in the clinic [50]. In a pilot study of ten patients with metastatic colorectal cancer a reduction in native T1 was seen in tumours treated with an anti-VEGF monoclonal antibody (bevacizumab) compared to pretreatment values [58]. Another study of 27 patients with acute leukaemia showed longer native T1 of vertebral bone marrow measured at diagnosis, compared with age-matched controls, and showed a decrease in T1 with treatment in patients who went on to obtain remission but persistently long T1 in patients who did not achieve remission [59].

Current limitations and future perspectives

The measurement of native T1 in multicentre trials is limited because good B0 and B1 homogeneity may be difficult to achieve, particularly over large fields of view or at higher field strengths. There is not yet a standard method of acquisition or analysis of data for estimation of native T1 as a biomarker in oncology. Future studies are required to investigate treatment-induced changes in T1, assess reproducibility of T1 measurements and validate observations against histological analysis. Accuracy, precision and reproducibility of T1 measurements can be assessed using phantoms containing materials of appropriate size, location and T1 [55].

Vascular biomarkers

Dynamic contrast-enhanced MRI

Mechanisms of contrast

DCE MRI provides information about the structure and function of the microvasculature [50, 55]. The term DCE MRI is used to refer to T1-weighted DCE imaging, whereas the term dynamic susceptibility-contrast (DSC MRI) is used to refer to T2*-weighted or T2-weighted DCE imaging [60]. DCE MRI uses a bolus injection of a paramagnetic gadolinium-based contrast agent to reduce the T1 of nearby protons and increase the signal intensity on a T1-weighted image. Estimation of T1 at each spatial and temporal position, combined with estimates of T1 before injection and knowledge of the relaxivity of the contrast agent, allows estimation of the concentration of gadolinium in each voxel over time [55]. Analysis of the gadolinium concentration–time curve using pharmacokinetic models or model-free approaches allows estimation of parameters related to the delivery of contrast agent to the tumour volume, the surface area and permeability of capillaries, the volume of the extracellular extravascular space (EES) and the blood plasma volume [55].

Technical requirements

DCE MRI requires rapid T1-weighted imaging before injection of a contrast agent, during injection and for several minutes (>5 min) after injection. Three-dimensional spoiled gradient-echo sequences are usually employed to achieve good temporal resolution (approximately 5 – 20 s) [55, 61, 62]. T1 can be estimated from two gradient-echo images acquired with different flip angles [53]. The larger flip angle is often acquired after injection of the contrast agent and combined with a low-flip-angle precontrast image to allow more rapid imaging. A power injector is usually used to provide reproducible administration of the contrast agent [55].

Pharmacokinetic models are often used to analyse the gadolinium concentration–time curve [60, 63, 64]. Off-line analysis is required using commercial or in-house software. Published recommendations state that the Tofts model, or equivalent, should be used and that the initial area under the gadolinium concentration–time curve (IAUGC) and the volume transfer constant (Ktrans) should be reported as primary end-points [55]. Other parameters, for example volume of EES per unit volume of tissue (ve) and blood plasma volume per unit volume of tissue (vp), are also reported in some studies, as well as model-free parameters, for example enhancing fraction [50].

Pharmacokinetic models require estimation of the arterial input function (AIF) which may be directly estimated from the DCE MRI data using an artery present in the images, or estimated using an additional sequence before the main DCE sequence using a prebolus of gadolinium; population-based estimates of AIF have also been applied [55].

Validation, qualification and clinical applications

Consensus recommendations have been published for the use of DCE MRI in early-stage clinical trials of antivascular and antiangiogenic therapies [55]. Recommendations for standardization of nomenclature in DCE MRI models have been produced [64]. Repeated baseline measurements in clinical studies have estimated the wCV of Ktrans and IAUGC to be around 15 – 20 %. [65, 66]. On comparison with histology, correlations between DCE MRI parameters and microvessel density have been demonstrated [62].

Changes in DCE MRI parameters with treatment, for example reduction in Ktrans but also other changes, have been reported in a variety of tumours (Figs. 1 and 5) and summarized in several reviews [50, 61, 62, 67]. A large number of these trials have been in a phase I setting with novel antiangiogenic agents [68–70] and have been particularly useful in paediatrics where radiation dose is a consideration [71]. In addition to its role in monitoring treatment response, there is an extensive literature on the utility of DCE MRI to predict patient outcomes [72–74]; for example, enhancement patterns predict overall survival (OS) in women with breast cancer undergoing neoadjuvant chemotherapy [75] and in patients with renal cell cancer [76].

Fig. 5
figure 5

Ktrans map of a large locally invasive breast cancer demonstrating radiological response. a Pretreatment image. b Posttreatment image shows a reduction in enhancement after two cycles of epirubicin and cyclophosphamide

Full size image

Current limitations and future perspectives

Implementation in multicentre trials is challenging as there are no standardized protocols for DCE MRI data acquisition [50, 55]. Analysis is done off-line and a range of models and software packages are available. Parameters derived from pharmacokinetic models depend on the choice of AIF [50, 55, 67]. The choice of ROIs also affects the results, and the methods for defining these are not standardized [55, 67]. Appropriate strategies for assessing heterogeneity and motion are also topics of current research [61].

Modelled parameters are dependent on many physiological processes, and interpretation is not straightforward [61]. The optimal timing of examinations relative to treatment is also unknown and DCE MRI may fail to detect an effect if there are rapid changes followed by a return to baseline properties after treatment [50]. Repeatability of DCE MRI parameters has been reported to be 15 – 20 %, which limits the size of change that can be detected. Changes in DCE MRI parameters found in phase I/II trials have not always translated into significant differences in progression-free survival or OS in phase III trials [67]. Going forward, standardization of acquisition and analysis methodology is crucial to successfully incorporate this biomarker as a robust imaging read-out in multicentre response assessment trials [34, 50].

Dynamic contrast-enhanced CT

Mechanisms of contrast

DCE CT provides information about tumour vasculature using rapid imaging before, during and after injection of an iodinated contrast agent. Pharmacokinetic modelling is used to estimate physiologically based parameters, for example blood flow per unit volume or mass of tissue (regional blood flow, BF), fraction of tissue that consists of flowing blood (regional blood volume, BV), time for the contrast agent to traverse the vasculature (mean transit time), rate of transfer of contrast agent from intravascular to extravascular space (blood flow extraction product, FE product) and permeability and surface area of capillary endothelium (permeability surface area product; Fig. 6) [77].

Fig. 6
figure 6

Maps of a blood flow and b blood volume calculated from DCE CT

Full size image

Technical requirements

Consensus guidelines recommend multislice imaging with high temporal resolution (about 2 s) [77]. Processing software is provided by the equipment manufacturers and is straightforward to acquire and robust to model because of the linear relationship between contrast agent uptake and increase in tissue density.

Validation, qualification and clinical applications

Consensus guidelines have been published for the use of DCE CT in oncology [77]. Repeatability studies have estimated wCV 16 % and 30 % for DCE CT parameters in preclinical [78] and clinical studies [66], respectively, which are similar to those for DCE MRI parameters. A multicentre study using a flow phantom showed minimal differences between DCE CT parameters among three institutions using a standardized protocol, which is a significant advantage for multicentre trials, but emphasizes that tube current and reconstruction methods could significantly affect results [79].

DCE CT vascular parameters have been shown to correlate with histological assessments of hypoxia [80] as well as with microvessel density [81]. DCE CT therefore has been used extensively in clinical trials where changes in BF, BV, FE product, and other parameters, have been used to demonstrate effects of various drugs in a variety of tumour sites [77, 82]. Key studies have been performed in non-small-cell lung cancer treated with sorefanib and erlotinib [83] or with radiotherapy [84], and in nasopharyngeal carcinoma treated with pazopanib [85] and cediranib in a phase I setting [66].

Current limitations and future perspectives

Radiation dose may be significant and should be kept as low as reasonably achievable whilst maintaining acceptable image quality [77]. This is a significant limitation in longitudinal studies. As with DCE MRI, software for analysis has not been standardized and there may be variation among results obtained from postprocessing software from different manufacturers [79, 86]. As with MRI, inclusion of DCE CT in multicentre studies requires standardization of acquisition protocols, QA procedures and analysis software [86].

Transverse relaxation rate

Mechanisms of contrast

Transverse relaxation rate (R2*), which is the reciprocal of the transverse relaxation time (R2* = 1/T2*), describes the rate of dephasing of transverse magnetization following excitation. R2* is determined by spin–spin interactions, inhomogeneities in the applied magnetic field (B0) and magnetic susceptibility variations in the tissue. In tumours R2* may reflect the presence of paramagnetic species, such as deoxyhaemoglobin, and may be related to oxygenation levels. It has also been suggested that a decrease in R2* in response to carbogen inhalation may reflect increased oxygenation in tumours which are hypoxic but have a functional vasculature [87–89]. An example R2* map is shown in Fig. 7.

Fig. 7
figure 7

R2* map acquired from an orthotopic PC3 prostate cancer xenograft

Full size image

Technical requirements

R2* can be estimated using a multiecho gradient-echo sequence. A log-linear plot of signal intensity against echo time has a gradient of −R2*. Good SNR, shimming and minimal motion between images are required.

Validation, qualification and clinical applications

Pre clinical studies have shown that baseline R2* and carbogen-induced changes in R2* estimates are correlated with histological assessments of hypoxia [88] and with tumour pO2 measured using a fibre-optic oxygen sensor [89]. A study carried out in a Th-MYCN genetically engineered mouse model of neuroblastoma showed slower baseline R2*, smaller change in R2* in response to 100 % oxygen and lower uptake of the perfusion marker Hoechst 33342 in tumours harbouring the ALKF1174L mutation, known to be associated with poorer prognosis in children with neuroblastoma, compared with tumours in the Th-MYCN cohort [90]. Pretreatment estimates of R2* and carbogen-induced ΔR2* have also been shown to be predictive of acute response to radiotherapy in two animal models, which the authors suggested may correspond to oxygenation levels of the tumours [91]. In assessment of response to chemotherapy, one preclinical study showed a decrease in R2* after treatment with a vascular disrupting agent (ZD6126) [92] while a study in another model showed increased R2* after treatment with an antiangiogenic agent (cediranib) but no change compared with controls in mice treated with either cyclophosphamide or a vascular disrupting agent (ZD6126) [56].

Only a small number of studies have demonstrated clinical applications of R2* as a biomarker in oncology. A study in breast cancer showed lower R2* in tumours than in normal breast tissue with an increase in R2* in tumours after neoadjuvant chemotherapy and a larger increase in responding patients than in nonresponding patients [93]. As with preclinical models, human studies have shown conflicting data; a correlation between R2* estimates in prostate tumours and measures of hypoxia determined by needle electrode oxygen measurements was demonstrated in one study [94] while another study in hepatocellular carcinoma showed that neither baseline R2* estimates nor R2* measured after inhalation of 100 % oxygen were able to detect microvascular invasion when compared to histopathological analysis [95]. As with DCE CT and MRI data, wCV of R2* in pelvic tumours is 17.5 % before treatment [65].

Current limitations and future perspectives

R2* has performed poorly compared with other imaging biomarkers. Changes in R2* between pretreatment and posttreatment measurements have performed worse than DCE MRI and DSC MRI in prediction of response to neoadjuvant chemotherapy in patients with breast adenocarcinoma [93]. It has been suggested that the increases and decreases in R2* estimates in response to treatment result from a combination of vascular effects; haemorrhage and other pathological changes in the tumour may contribute to observed changes in R2* and that estimates of R2* cannot discriminate between these changes [87, 92].

Vessel size index

Mechanisms of contrast

Vessel size index (VSI) is the average diameters of blood vessels within a voxel. Injection of a paramagnetic or superparamagnetic contrast agent causes different changes in R2 and R2*, and the ratio ΔR2*/ΔR2 has been shown to be related to vessel sizes in tumours and normal tissues [96]. Analytical expressions have been derived relating VSI to ΔR2*/ΔR2, the change in magnetic susceptibility of blood due to the contrast agent (Δχ) and ADC [97, 98]. An example VSI map is shown in Fig. 8.

Fig. 8
figure 8

VSI map acquired from a murine B16 melanoma xenograft

Full size image

Technical requirements

ΔR2 and ΔR2* can be measured using spin-echo and gradient-echo sequences, respectively. Pre clinical experiments have used steady-state experiments using ultrasmall paramagnetic iron oxide (USPIO) contrast agents, which are assumed to remain in the blood pool [96]. Estimates of Δχ can be obtained from ex vivo measurements [97]. ΔR2* and ΔR2 have also been estimated using first-pass dynamic sequences with gadolinium-based contrast agents in clinical and preclinical studies [99–101].

Validation, qualification and clinical applications

Although correlation has been shown between VSI estimated from MRI and histological analysis, MRI overestimates VSI compared to histology [102], two-photon laser scanning microscopy [103], intravital microscopy [104] and micro-CT [105]. Conversely, other studies have shown good agreement between VSI estimates from in vivo MRI and ex vivo measurements of vascular casts using micro-CT [106]. Pre clinical studies have shown increases in mean VSI in tumours treated with antivascular agents, attributed to loss of small functional vessels; histological analysis have shown similar relative changes despite the discrepancy in absolute values [107, 108]. Other studies, however, have shown a decrease in VSI after treatment [109].

Only a small number of clinical studies have used VSI in a clinical trial setting. One study in gliomas showed a correlation between ΔR2*/ΔR2 and tumour grade [99] whilst another study of 16 patients with glioblastoma used maps of relative tumour vessel size to show reduction in tumour vessel size in patients treated with a pan-VEGF receptor tyrosine kinase inhibitor as part of a phase II trial [100].

Current limitations and future perspectives

Early development of USPIOs for clinical use was halted but the use of ferumoxytol has recently been reported in DSC MRI [110]. Motion may also affect estimates of VSI, particularly in extracranial applications. Overestimation of VSI, due to simplifying assumptions in the model [98] or inability to detect vessels that are not perfused by the contrast agent [102], may limit VSI to relative rather than absolute measurements. Therefore, although relative changes in VSI have shown promise in preclinical measurements, validation of quantitative results is required.

Relative blood volume

Mechanisms of contrast

Relative blood volume (rBV) is the proportion of a voxel that is composed of blood. Injection of a paramagnetic or superparamagnetic contrast agent causes a change in R2* which is dependent on the vascular architecture. Analytical expressions have been derived relating rBV to ΔR2* and Δχ [97]. An example rBV map is shown in Fig. 9.

Fig. 9
figure 9

rBV map acquired from a 786-O RCC xenograft using susceptibility-contrast MRI with USPIO particles

Full size image

Technical requirements

R2* can be measured using gradient-echo sequences. In pre clinical experiments involving steady-state experiments USPIO contrast agents have been used which can be assumed to remain in the blood pool [97]. In clinical studies DSC MRI has been used with injection of gadolinium-chelate contrast agents or, more recently, the USPIO ferumoxytol [110].

Validation, qualification and clinical applications

Reductions in rBV have been shown in preclinical studies of tumours treated with antiangiogenic [107] and vascular disrupting agents [109] in good agreement with histology results, which have shown reduction in perfusion in the treated tumours. In vivo MRI has been shown to overestimate rBV compared with ex vivo micro-CT [105] or two-photon laser scanning microscopy [103]. Some preclinical studies have shown good agreement between estimates of rBV from MRI and histological measurements of blood volume [111] although other studies have shown that estimates of rBV from MRI are larger than estimates from histology [102].

In clinical studies, wCV of rBV in pelvic tumours was 19.7 %, similar to DCE MRI parameters [65]. rBV has been shown to be able to distinguish between types and grades of brain tumours [112] and to distinguish recurrence or progression from posttreatment radiation effects, necrosis and pseudoprogression [110, 113]. rBV has also been shown to be predictive of time to progression or OS in gliomas [114]. A small number of studies of rBV have been carried out in extracranial tumours [115]. One pilot study in ten patients with renal cell carcinoma showed a decrease in rBV after treatment with sunitinib [116]. A study in 37 patients with breast cancer treated with neoadjuvant chemotherapy showed that the change in rBV between pretreatment and posttreatment scans was correlated with clinical and pathological response [117]. A study of 20 patients with prostate cancer showed a decrease in rBV 1 month after starting androgen deprivation therapy [118].

Current limitations and future perspectives

Although rBV has been used in many clinical studies for the assessment of brain tumours, only a small number of studies have investigated extracranial tumours. As with other vascular biomarkers, standardized methods for acquisition or analysis of rBV estimates from DSC MRI or steady-state measurements are lacking, which precludes implementation of this biomarker in a multicentre setting. As applications in brain tumours have far exceeded extracranial applications for this biomarker, attempts at standardization in brain protocols is currently underway. Further biological validation, however, is urgently needed [115].

Metabolic biomarkers

Magnetic resonance spectroscopy

Magnetic resonance spectroscopy (MRS) uses largely the same hardware as MRI, but instead of acquiring high-resolution images of water and lipid distribution and their properties, it acquires signals from compounds of low molecular weight in tissue that have concentrations of a few millimoles. MRS can therefore probe biochemistry and metabolism in tissue. In addition to tissue characterization, it can also be used to help evaluate response and recurrence, and aid treatment planning. Example 1H MR spectra are shown in Fig. 10.

Fig. 10
figure 10

Means and standard deviations (vertical lines) of normalized STEAM (echo time 30 ms) spectra: normal (parietal) white matter (N = 6), meningioma (N = 8), metastases (N = 6), astrocytoma grade II (N = 5), anaplastic astrocytoma (N = 7), glioblastoma (N = 13). Reproduced from Howe et al. [126]

Full size image

Technical requirements

Signals are acquired either from a single specified voxel or from a 2D or 3D array of voxels using some form of MRS imaging. The minimum useful voxel size depends on magnetic field strength, RF coil design, scan duration and the question being addressed, and is typically about 10 – 20 mm. Methods are well-described in the literature [119]. Manufacturers provide some software for spectral data processing, display and fitting, but many users will also process data off-line using specialized spectral processing packages such LCModel [120, 121] and jMRUI [122–124]. Results are commonly expressed as ratios of peak areas within the spectrum, or relative to signal from tissue water (for 1H MRS). Calculation of metabolite concentrations is possible if values for the T1 and T2 relaxation time constants of the metabolites are known, and where the RF transmit and receive fields are either uniform or amenable to calculation. To acquire MRS signals from magnetic nuclei other than 1H requires a RF system (coils, amplifiers, detectors etc.) that operate at the appropriate MR frequency.

Proton MRS and its clinical applications

Most MRS studies use signals acquired from 1H nuclei of compounds in tissue, since 1H nuclei provide the largest signals, and this requires no hardware modification to the scanner. 1H MR spectra of normal brain are dominated by choline, creatine, and N-acetyl aspartate (Fig. 10). Brain tumours are characterized by reduced N-acetyl aspartate, and often by elevated total choline (which includes choline, phosphocholine and glycerophosphocholine), lipid or lactate. Elevated choline is attributed to increased proliferation and demand for membrane synthesis, while lactate may be caused by a combination of increased lactate production owing to the Warburg effect [125] together with inadequate perfusion to remove it. Different brain tumours have different metabolic fingerprints, yielding the possibility of using 1H MRS for differential diagnosis [126] (Fig. 10). Regions of metabolic abnormality sometimes extend beyond regions of imaging abnormality [127], demonstrating that MRS can detect regions of tumour not detected by standard MRI.

Elevated choline is characteristic of tumours in other tissues also, such as breast [128] and prostate [129]. Prostate tumours also have reduced citrate and spermine [129]. Some tumours exhibit high levels of lipids, in particular high-grade glioma and metastases in the brain [126]. These signals arise from cytoplasmic lipid droplets rather than from the membranes of cells and organelles, as lipids in bilayers are relatively immobile and produce signals that are too broad to be detected using standard MRS methods. They are associated with proliferation, inflammation, malignancy, necrosis and apoptosis [130, 131]. In tissues such as breast and prostate care is required to ensure that signals acquired are not contaminated by those of surrounding lipid.

Nonproton MRS and its clinical applications

31P MRS gives lower signals than 1H MRS but is useful for probing energy metabolism (phosphocreatine, ATP, NADH) and some compounds involved in membrane synthesis and breakdown, in particular phosphomonoesters (PME) such as phosphoryl choline, and phosphodiesters (PDE) such as glycerophosphorylcholine. Cancers tend to be characterized by elevated PME and PDE, but relatively normal ATP. Many studies have demonstrated a reduction in PME in response to treatment [132, 133]. In non-Hodgkin’s lymphoma a multicentre trial has also demonstrated that 31P MR spectra acquired before treatment contain prognostic information; tumours with an initially lower PME/NTP ratio are more likely to respond than those with a high PME/NTP [134]. 31P MRS can also be used to measure intracellular pH [135, 136]. Applying this method to tumours has yielded the surprising result that tumours generally tend to maintain a slightly alkaline intracellular pH in spite of the Warburg effect [137]. Measurement of intracellular pH in vivo would be useful in assessing the effects of anticancer strategies that are anticipated to alter pH, such as inhibitors of monocarboxylate transporters [138].

19F MRS yields signals almost as strong as 1H MRS. While there is little MR-visible 19F in the body, 19F MRS has been used to follow the distribution and metabolism of anticancer drugs such as 5-fluorouracil [139] and to study hypoxia using perfluorocarbons [140] and fluorinated nitroimidazoles [141].

Current limitations and future perspectives

Magnetic resonance spectroscopy has great potential for probing metabolism and biochemistry of tissues in vivo, but is often limited by the relatively low SNR. Dynamic nuclear polarization is a new method in which atomic nuclei of compounds with large longitudinal relaxation time constants can be prepolarized to yield signals 10,000-fold larger than the normal polarization. The first-in-human trial of this method has recently demonstrated potential for imaging highly elevated pyruvate-to-lactate conversion in prostate tumours [142]. However, the full potential of this technique in providing information on metabolic pathways and thus monitoring enzyme kinetics remains to be exploited.

Conclusion

A range of biomarkers providing information on tissue organization, vascular properties and metabolism are available with MRI supported by vascular biomarkers with CT. Their potential for delivering information not only for response assessment, which has traditionally been the case, but also to predict disease aggressiveness, treatment response and outcomes of therapy offers rich avenues for exploration. Even in response assessment, it is expected that these biomarkers will deliver information on how the tumour is responding much earlier than has hitherto been possible, thus sparing the patient the morbidity of ineffective therapy and the opportunity to switch to a more effective treatment regimen earlier. As we uncover information on the heterogeneity of these biomarkers, it will also become clear which biomarker combinations are most informative, as it is unlikely that a single parameter will contain the depth of information required. Finally, in order to exploit these biomarkers fully in large multicentre trials, it is imperative that we achieve standardization with consensus on acquisition and analysis protocols that optimizes reproducibility of the measurements and allows pooling of multisite data.