Medical imaging in personalised medicine: a white paper of the research committee of the European Society of Radiology (ESR)

Screening can be considered as PM as it can provide treatment of disease in a timely fashion before the disease has become clinically manifest with a decreased chance of implementing curative treatment. Imaging-based screening programmes are not only performed in healthy subjects. Patients can also enter an imaging screening programme to detect the progression of disease or development of cancer in an earlier phase (Table 3). The objective of these screening programmes is to start treatment earlier than it would be if based on clinical progression. Well-known examples are the image-based screening programmes for breast, lung, colorectal and prostate cancer [16‐18]. Some programmes include all subjects within a certain age range; others identify high-risk categories based on simple algorithms.

The main problem with the detection of asymptomatic disease is the potential for overdiagnosis. Overdiagnosis is the diagnosis of irrelevant diseases or diseases that are sufficiently stable or indolent that they would not have become clinically relevant during the subject’s lifetime. Overdiagnosis leads to unneeded treatment with potentially harmful side effects and thus causes both economic and emotional burden. The magnitude of overdiagnosis has been recently estimated to be as much as 25 %–60 % in breast, lung and prostate cancer screening [19]. Fortunately, imaging not only plays a role in the detection of subclinical abnormalities, but can also be used to characterise the lesion as aggressive or indolent. For example, the large number of non-specific nodules detected during lung cancer screening can be addressed by follow-up imaging to assess the change in the volume of the nodule, thereby reducing the number of false-positive diagnoses [20]. A complementary molecular imaging technique allows for better characterisation of incidental findings from screening procedures, such as solitary lung nodules, dense breast lesions and suspected prostate cancer, thus avoiding unnecessary invasive procedures and biopsies in patients with negative molecular scans [21].

Currently, prevention of cardiovascular disease and stratification of the healthy population are based on the Framingham risk score or Systematic Coronary Risk Evaluation (SCORE) score. Based on classical risk factors including age, gender, systolic blood pressure, hypertension treatment, smoking, total and HDL cholesterol, an individual is classified as having a low, intermediate or high risk. High-risk individuals are treated with intensive medical treatment while low-risk individuals are helped with lifestyle modifications. The main problem here is the large intermediate-risk group. Extensive research has been performed to better stratify patients within the intermediate risk group. As imaging is able to detect presymptomatic atherosclerotic disease, it is logical to hypothesise that imaging may become the key to identifying people who could benefit from preventive intervention. Candidate imaging biomarkers are the CT-derived calcium score from the coronary arteries [22] or more peripherally located arteries, US-derived intima-media thickness, US- or MRI-derived pulse wave velocity [23], or MRI-based plaque composition [24]. These imaging biomarkers can predict later development of disease, either on their own or by supplementing established risk factors [22]. The development of such new prevention strategies requires imaging at the population level. Population imaging is performed within the context of large population-based, prospective epidemiological studies. Population imaging focusses on finding imaging biomarkers that allow early diagnosis of preclinical disease and prediction of clinical disease. Besides imaging, information about the history of the individual subject, blood and body fluid samples, genetic data and most importantly follow-up data with regard to the development of symptoms and disease are available for correlation.

Although the question “Where is the disease and what is it?” may have been clearly answered, a clear decision on the optimal treatment cannot always be made without taking into consideration additional imaging data. For example, knowledge of the extent of a disease—a parameter of the severity of disease—is crucial for decisions on the choice of treatment. The classical example is tumour staging in which the involvement of adjacent tissues, regional lymph nodes and distant metastasis has an immediate and important impact on prognostic stratification and on the choice of therapeutic options. Improved staging with the use of imaging might help in avoiding unnecessary surgical interventions, which increases the morbidity of disease without a substantial benefit in the reduction of mortality. On the other hand upstaging can lead to the choice of more aggressive therapies, which may reduce the recurrence rate [25, 26].

Tumour staging relies strongly on cross-sectional imaging, with CT and/or MRI information being combined with quantitative metabolic information from PET and SPECT imaging. General consensus criteria and guidelines have been established based on large clinical studies combining multi-modality imaging criteria [27‐29]. In these guidelines PET/CT is increasingly used in staging. This technique, which visualises specific tracer uptake in normal and abnormal tissue, allows for better localisation of cancer in structurally normal tissue. Moreover, the whole-body approach increases the sensitivity for distant metastases. The improvement in staging with PET/CT compared to PET or CT alone is substantial and has been reported for many cancers and cancers of unknown primary [30, 31]. Cost-effectiveness of preoperative PET has been proven with, for example, prevention of unnecessary surgery in one out of five patients with suspected non-small-cell lung cancer after the addition of [¹⁸F]FDG-PET imaging to conventional staging [32]. Recent emergence of combined PET/MR imaging systems is expected to raise the potential for combined criteria in better assessment and tissue characterisation of specific tumour and pathological lesions.

Besides oncology, cardiovascular and cerebrovascular diseases are other areas where severity of disease influences treatment decisions. Specific examples are:

Final diagnosis of specific diseases and subsequent treatment allocation has been based on gross macroscopic, microscopic and/or immunohistochemical evaluation of tissue specimens. The best-known example is a tumour for which the most relevant question is the assessment of the malignant nature of the lesion and the type of malignancy. Collection of tissue samples for diagnosis can been performed by endoscopy or with a surgical procedure. US-, CT- or MRI-guided biopsy sampling is critical and most often represents the least invasive method to obtain relevant material for the in vitro diagnosis of disease in many parts of the body. Specific examples include:

Tissue sampling is not only performed for a final pathologic diagnosis, but also for the evaluation of disease spread. Here, the confirmation of potential metastasis through image-guided tissue sampling may change the type of treatment offered, for example from surgery to radiotherapy and/or chemotherapy. Specific examples include:

The essence of personalised diagnosis for personalised treatment lies in the use of biomarkers. These can be extracted from tissue, body fluids or medical imaging. In the selection of the right treatment, two different types of biomarkers are of particular importance: prognostic and predictive biomarkers. Prognostic imaging biomarkers predict the likelihood of disease progression in the absence of treatment considerations. Such biomarkers are valuable not only for identifying aggressive disease that requires immediate treatment, but also for identifying indolent disease that can safely be left alone.

Predictive biomarkers indicate the relative sensitivity or resistance to drug treatment or the success of treatment by surgery, interventional radiology or radiotherapy.

Both types of biomarkers are relevant in PM as they influence the decision for the treatment, type, length and intensity of treatment. A specific application of a prognostic or predictive biomarker is the monitoring of the response after the first cycle of treatment (chemotherapy and/or radiotherapy) in order to predict the final response, progression-free survival, recurrence and overall survival.

Multiple prognostic and predictive imaging biomarkers have been identified. The main challenge for PM is the translation of this information in patient management pathways and treatment algorithms.

PM implies the identification of the specific genomic and molecular substrate of alterations leading to disease and “medical treatment tailored not just to symptoms, but to the biochemical profile of an individual’s disease state (gene expression, proteome, dominant metabolic pathways, etc.)” [48]. This has tremendously changed the evaluation of diseased tissue specimens: from microscopic and/or immunohistochemical evaluation to gene expression profiling.

Parallel to this development, medical imaging has changed. Traditionally, medical imaging has been a tool for non-invasive assessment of anatomy and for detection and localisation of disease processes. Currently, a wide variety of new medical imaging techniques produces important information about physiology, metabolism, molecular biological processes and functional genomics. Multiple methods have been described for imaging the temporal and spatial biodistribution of a molecular probe as well as related biological processes such as cell proliferation, apoptosis, angiogenesis, hypoxia, and gene activation and expression. These new methods combine the ability to measure and quantify biological processes with the ability to localise the measured entities onto a high-resolution anatomical and functional image [49]. In the future, an imaging diagnosis will not only include the location and extent of diseased lesions but also information on the physiological characteristics and biological behaviour of the lesion (e.g., perfusion, flow, metabolism, diffusion).

Molecular imaging is likely to replace several aspects of ex vivo tissue analysis, which will be a key component in PM. It is expected that molecular imaging can be used to detect diseases much earlier because molecular and cellular changes precede anatomical and structural changes on classical imaging modalities such as CT and MRI. As such, it will contribute to PM by treating the patient at the right time. As new therapies will be increasingly based on a foundation of precise understanding of the molecular basis of disease, simultaneous creative probe development will require a deeper understanding of cancer biology [50].

PM currently requires an in-depth analysis of the genetic expression and molecular and cellular biology of the individual disease. This is based on a more refined evaluation of tissue samples by assessment of molecular and genetic expression patterns. For example, for malignant tumours it is has become relevant to know which of the genetic alterations is functionally relevant for abnormal cell proliferation.

Recent studies have shown that malignant lesions demonstrate intratumoral heterogeneity with regard to genetic alterations and expression patterns, thus indicating the existence of multiple subpopulations of cells. Furthermore, metastatic lesions may have genetic expression patterns that differ from the expression pattern in the primary lesion or from each other, and these lesions can therefore show a different behaviour and response to treatment. For more effective targeted therapy, it is therefore necessary to identify these mutations and confirm them by individual tissue analysis of each subset.

The differences in genetic expression patterns are frequently reflected as differences in imaging appearance. Imaging can thus further play a crucial role in guiding the sampling procedure. For example:

One of the classical examples of PM is the determination of the status of the human epidermal growth factor receptor 2 (HER2) expression in patients with breast cancer as a prerequisite for treatment with trastuzumab (Herceptin, Genentech). In some cancers, HER2 is over-expressed, causing cancer cells to reproduce uncontrollably [58]. Trastuzumab is a monoclonal antibody that interferes with the HER2 receptor and inhibits the effects of overexpression of HER2. This approach targets one of the core deregulated cellular signalling pathways in cancer, which is totally different from cytotoxic chemotherapy. If the breast cancer does not overexpress HER2, trastuzumab will have no beneficial effect. The coupling of therapy with diagnostic information specific for the intended target is called a theranostic system.

In the HER2-trastuzumab theranostic system, the diagnostic test is a laboratory test on a tissue specimen in vitro, but in vivo imaging can also provide diagnostic tests for theranostic systems. The most common example is ¹²³I/¹³¹I for targeted radionuclide radiotherapy in differentiated thyroid cancer. Targeted radionuclide therapy combines “the favourable targeting properties of peptides and antibodies with the effectiveness of radiation-induced cell death. A major advantage is the possibility to determine the selective accumulation in the targeted tissue by molecular imaging studies using structurally identical diagnostic compounds” [59, 60].

Theranostics relying on in vitro testing of small tumour samples does not address the problem of tumour heterogeneity or the phenotypic dedifferentiation of tumour metastases. By using different molecular imaging tracers, we begin to appreciate the heterogeneity of metastatic disease within individual patients. Tumour heterogeneity can often be seen even within the same lesion. Thus, in a patient with metastases of differing biology, a single therapy approach may result in a mixed tumour response. It is not possible to biopsy each and every lesion, so molecular imaging is essential to successfully tailor treatment to the individual tumour biology. A theranostic concept that uses molecular imaging will better address the complexity of tumour biology and will therefore have the potential to enable improved diagnosis, treatment, and monitoring of treatment response all at once [50].

Theranostics provide a unique signature for improved staging of the tumour, for imaging of the biodistribution of the target to predict the biodistribution of the radiation dose and to individually monitor the efficacy of treatment with the same basic compound that is being used to target and treat the disease [60]. Furthermore, the development of new specific biomarkers that can be used for both in-vivo diagnostic imaging and targeted therapy with alpha emitter compounds or carriers of cytotoxic agents will further expand the scope of the application of theranostics in clinical practice. It is even foreseeable that advanced targeted therapy, which is often extremely expensive, would only be applicable if we have a companion diagnostic biomarker that can predict the efficacy of the treatment.

Aside from the targeted therapy with ¹³¹I, other examples of image-based theranostics are:

The evolving era of nanotechnology will further boost the development and use of theranostics. Nanoparticle-based imaging and therapeutic agents are a specific, advanced and promising form of theranostics. Nanoparticles are synthesised from organic or inorganic materials such as lipids, polymers, metals or semiconductors. Thereafter ligands have to be conjugated to the nanoparticle in order to target cellular receptors (e.g. overexpressed in tumour-induced angiogenesis) or other molecular characteristics of diseased tissue [64]. In addition, visualisation is achieved by conjugation of imaging agents such as radionuclides or paramagnetic/superparamagnetic metals. If the diagnostic scan demonstrates that the lesion is positive for the target, the patient can be treated with the same nanoparticle, which now carries the therapeutic agent. The therapeutic agent can be cytotoxic radionuclides or drugs encapsulated in the nanoparticle structure. In the former case, the nanoparticles have to be engineered with a moiety for chelation of radiometals. Other materials that can be delivered locally are interfering RNAs, genes and peptides. A major advantage of nanoparticles for theranostics would be the high payload of both the therapeutic agent and imaging probe that can be delivered to the target tissue. Specific examples of theranostic agents currently under investigation include:

Radiogenomics is a new approach in which large sets of complex descriptors of disease are extracted from routine clinical images and related to molecular biology and gene expression patterns of disease [70, 71]. It is based on the assumption that the observed image is the product of mechanisms occurring at a genetic and molecular level. Thus, parameters extracted through image processing and analysis are linked to the genotypic and phenotypic characteristics of the tissue [72].

Like other ‘omics techniques, it is not hypothesis-oriented but goal-oriented. In radiogenomics, an a priori hypothesis is not formulated on which imaging technique or feature should be used to reflect a certain biological mechanism. Every imaginable feature is extracted from an image data set, including semantic descriptors (e.g., shape, spiculated patterns, presence of necrosis, localisation) and/or quantitative features. These features are then correlated to gene expression profiles. Radiogenomics is the term used when imaging features are correlated to gene expression. The very large numbers of correlated parameters (usually over a hundred, ideally several hundreds) extracted from the same data sets requires specific statistical methods similar to those used in other ‘omics.

One of the first radiogenomics publications linked CT imaging features to gene expression in hepatocellular carcinoma [73]. The expression variation of over 6,000 genes was grouped into 116 gene modules. By combining 28 imaging traits into similar modules, 78 % of gene expression profiles could be predicted. This means that the relative expression of a given gene could be determined in any given liver tumour. This approach has been successfully applied to MRI features of glioblastoma multiforme and gene expression profiles, protein expression [74, 75], microRNAs [75], messenger RNA expression and DNA copy number variation [76]; to MRI features of breast cancer and gene expression profiles [77]; to CT features of renal cell carcinoma and gene mutations [78] and to [¹⁸F]FDG features of lung cancer and protein expression [79].

It is clear that medical images inherently contain a wealth of mineable information, which reflects genetic and molecular information of disease in an individual patient. This would potentially allow particular imaging phenotypes to serve as surrogates for the unique molecular programmes that typify a molecular subtype of cancer [80]. Radiogenomics will definitely contribute to personalisation of patient management and help improve existing staging and diagnostic schemes. Imaging can be used to determine the molecular diversity of disease and thereby stratify patients into molecular subclasses with different prognoses [72]. For example, one study demonstrated that imaging features of hepatocellular carcinoma were predictive of a gene expression profile of venous invasion genes and subsequently predictive of a poor outcome [73]. Equally important is the use of imaging to identify patients for whom targeted therapeutic agents will be effective. For example, one study demonstrated that specific imaging traits in GBM were associated with an EGFR protein expression programme and EGFR protein expression [74]. Finally, intratumoral genetic heterogeneity can be reflected in intratumoral imaging heterogeneity, which demonstrates one advantage of imaging over tissue analysis.

Although PM has been boosted by the huge developments in molecular profiling, it is clear that nowadays imaging can provide comparable information. However, competition between the two approaches is not foreseen as each acts on a different level. Michael Kuo, one of the pioneers in radiogenomics, sees a blurring of the traditional lines between the diagnostic sciences and a merging of molecular diagnostics and imaging into a single diagnostic discipline [80]. Integrated diagnostics must include multimodality and multiparametric imaging approaches as well as multiple pathologic, molecular and genetic assays. This integration of the huge diagnostic data from all levels—from genes, molecules, cells, tissues, organs and organ systems—is needed in the era of PM.

Treatment monitoring in malignant tumours is based mainly on anatomical imaging, frequently still assessed on 2D images, e.g., according to the RECIST criteria. These criteria have been expanded (RECIST 1.1) to include lymph node evaluation and the use of PET [82]. Anatomical imaging detects morphologic changes, which occur relatively late during treatment. Over the last decade, some shortcomings of anatomical imaging with regards to therapy response have become evident [83]. These relate to a poor reproducibility of tumour measurements [84, 85] and inconsistencies between tumour response and survival [86]. Anatomical imaging alone does not support a comprehensive understanding of tumour biology. Furthermore, anatomical imaging falls short of fully assessing the response to targeted therapies that exert a cytostatic rather than a cytotoxic effect as well as the early response assessment to neoadjuvant systemic treatments.

The field of treatment response evaluation is a prime example of the gap between current practice and advances in imaging and image analysis techniques. Three-dimensional display of lesions and volumetric measurements have advanced and may yield advantages in the assessment of tumour response. For example, comparison of both single-diameter and volumetric measurements to assess treatment response demonstrated significant differences in tumour response [87, 88]. In addition, volumetric measurements for response assessment provide more reproducible results [89]. Advances in quantitative imaging and image analysis will contribute to changing the landscape of clinical practice and clinical trials [90, 91].

The development of new imaging techniques such as molecular, metabolic and functional imaging has paralleled the expanding knowledge on tumour biology. These techniques provide additional critical information for care of individual patients when used as a supplement to anatomic imaging, as they expand the ability to biologically characterise tumours and monitor tumour response at a more appropriate level. Cancer is currently seen as a dysregulation of a restricted number of signalling pathways, and treatments that target these pathways will lead to successful therapies. Molecular imaging helps assess changes in the tumour that would indicate the activity of one or more of the key dysregulated pathways. For example, for quantitative analysis of tumour response, [¹⁸F]FDG-PET imaging — which assesses the glucose uptake and thus the metabolic activity level of various cancers — has been identified as a reliable indicator of treatment response. PET-based scoring criteria extend the morphological RECIST criteria into a biological dimension with a proposed PERSIST score [92]. This approach was found to be more consistent and showed better prediction of patient response to the treatment in a variety of cancers. A wide range of more specific probes and new imaging biomarkers are currently emerging for assessment of individual types of cancer and monitoring response to treatment.

Advanced imaging can also target the problems related to the use of one single marker for treatment response evaluation. While intratumoral heterogeneity explains the individual response to (targeted) therapy, a monitoring approach based on a single marker is not sensitive to non-response in a heterogeneous lesion or in one of multiple lesions. “This defines the weakness of serum analyses, which provide an average signal of output from all lesions, or of a biopsy program to characterize a tumour—if all can be different, then all must be characterized” [93]. Accurate localisation of regional differences in tumour behaviour and tissue characteristics is the strength of imaging. But even a molecular imaging approach based on a single probe may be insufficient, as alternative dysregulated signal pathways may replace the function of the targeted pathway [93, 94]. Currently, imaging can be performed with different modalities integrated into hybrid systems (SPECT/CT, PET/CT or PET/MRI) and can cover several functional levels. Multi-parametric imaging can potentially better identify heterogeneous and localised response to treatment. By combining imaging markers from different pathways, it may be possible to assess the type of physiological escape mechanisms of a tumour (or other disease) [94, 95], enabling individual treatment to be fitted in real time to the type of response which is, of course, PM in its most basic sense.

Treatment response is relevant for clinical practice but equally important for pharmacological research, where it can be used to identify and evaluate novel drugs. Indeed, in its Critical Path Initiative, the US Food and Drug Administration (FDA) has recognised the new role of medical imaging in drug development and regulatory approval [97]. Although the Initiative emphasises the role of imaging in the assessment of biomarkers, medical imaging also has other applications in drug discovery: “a molecular imaging probe can be used to determine target occupancy before and after administration of a new drug candidate, which helps to assess binding affinity and to determine correct dosages” [98].

In pharmaceutical research, the demonstration of tumour response is an important element of early drug development. Further investigation is often not performed when antitumour activity and treatment response are not demonstrated. Anatomical imaging is limited to revealing a physical shrinkage of the tumour. With the current molecular and functional imaging techniques all kinds of antitumour activity can be detected such as changes in receptor expression and tumour perfusion. It is expected that these new techniques, which integrate knowledge on the tumour biology, will foster the detection of relevant targeted drugs.

Very important is the increasing use of (imaging) biomarkers as surrogate endpoints for clinical trials. Given that validated surrogate endpoints allow dramatic shortening of clinical trials by shortening the time to the study end point, with associated savings and a faster availability of new treatments, this is an area in which both the pharmaceutical industry and healthcare authorities are pushing for biomarker use. Molecular imaging and radiopharmaceuticals are particularly important within this context as they allow in vivo visualisation of the effect of the treatment.

Imaging has always been involved in radiotherapy through image-guided radiation therapy planning (RTP), including the verification of treatment delivery. Delineation of primary and metastatic lesions, which helps to optimise target volume definition in radiotherapy, is based on imaging. Initially, the introduction of cross-sectional imaging has tremendously improved the field of radiation oncology [100]. Intensity-modulated radiotherapy, which improves dose delivery to the tumour while reducing exposure of healthy tissue to limit the side effects of treatment, requires a precise delineation of the tumour. For that reason MRI is increasingly used in treatment planning as it provides better soft tissue contrast, which improves target volume definition. A next step is the application of functional imaging techniques, which can provide information on more aggressive parts or more radiation-resistant parts of the tumour. The target volume definition is then based on the biology of the tumour.

Some tumours are located in areas of the body that are prone to movement, such as the lungs, liver and prostate gland, and some tumours are located close to critical organs and tissues. In these patients, image-guided radiation therapy (IGRT) is used in which imaging immediately before or even during a course of radiation therapy is applied to improve the precision and accuracy of the delivery of treatment. In addition, another application of IGRT is the reduction of the target volume during the course of radiation treatment based on the visualisation of tumour shrinkage with imaging.

Image-guided interventions provide the means to deliver therapies locally at the disease site, whether the therapies are based on chemical compounds (including drugs and radiotherapeutics), genes, devices (including stents), cells, sound waves (high-intensity focused ultrasound = HIFU) or induction of ablative temperatures (using hyper- and hypothermia with radiofrequency, microwaves or cryotherapy). Most current interventional image-guided procedures are individually tailored to the local anatomic and functional circumstances as well as the personal needs of the patient. In this way, image-guided interventions are in themselves an important and integral part of PM. For example:

Medical imaging is going to play an important role in individualised drug delivery. In the past 30 years, drug delivery research was focussed on targeting carriers such as liposomes, polymeric nanoparticles and micelles to the disease site. New developments now promote the selective release of bioactive compounds after application of a trigger [102]. The trigger mechanism can be applied locally and at specific times, depending on the individual patient’s needs, and can be selectively applied with interventional techniques such as HIFU under the guidance of MRI [102]. Although for many reasons most tumour ablations are performed using percutaneous radiofrequency or microwaves, HIFU is the only clinically viable technology that can be used to achieve a local temperature increase deep inside the human body in a non-invasive way. MRI can be used to provide continuous temperature mapping during HIFU for spatial and temporal control of the heating procedure and prediction of the final lesion based on the received thermal dose [103].