Dynamic contrast-enhanced magnetic resonance (DCE-MRI) is a functional imaging technique. It consists of MRI scans coupled to the injection of a contrast agent. The latter leads to a decrease in relaxation time and provides extremely detailed characteristics of the micro-circulation of blood through tissue.
DCE-MRI assessments typically use the characteristics of signal intensity (SI) and time-intensity curves (TIC) regarding regions of interest (ROI). Early DCE-MRI efforts assumed a linear relationship between signal enhancement and contrast uptake. However, given that signal enhancement depends to a very great degree on intrinsic tissue and acquisition parameters, more complex models have been developed to control the effect of tissue characteristics such as the pre-contrast longitudinal relaxation time and the longitudinal or transverse relaxivities of the contrast agent.

Two-phased process
DCE-MRI is a two-phased process. Typically, at first, a T1-weighted MRI scan is conducted. This is followed by injection of the contrast agent, and then repeated acquisition of T1-weighted fast spoiled gradient-echo MRI sequences to obtain measurements of signal enhancement as a function of time.
The contrast agents are usually based on gadolinium and include gadoterate meglumine (Gd-DOTA), gadobutrol (Gd-BT-DO3A) gadoteriol and albumin-labelled Gd-DTPA.

Image acquisition and voxel comparison
Typically, 3D image sets are obtained sequentially every few seconds for up to 5-10 minutes. Shorter intervals allow for detection of early enhancement, although many researchers consider 10 seconds to be good enough. Longer intervals than this typically makes it tougher to identify early enhancement.
At the moment, the debate about the upper limit for intervals continues.
After image acquisition, the comparison of T1 values per voxel in each scan allows identification of permeable blood vessels and tumour tissue. Both spatial and temporal resolution must be adjusted to obtain an adequate sampling of the contrast enhancement over time, for each tissue voxel. The speed with which MRI images must be acquired necessitates larger voxels, so as to maintain adequate signal-to-noise ratios. Thus, DCE-MRI is often not as high in resolution as conventional T2-weighted sequences.

Range of biomarkers
Although DCE-MRI can be performed on conventional scanners (typically 1.5 T), it requires specialist image analysis to analyse the enhanced biomarker information which is to be provided. Such information includes tissue perfusion, vascularity, endothelial permeability, cellularity etc.
The biomarkers can be used to provide measurements of tumour vascular function and to improve the diagnosis and management of diseases in a variety of organs.

DCE-MRI in the brain
Clinical applications of DCE-MRI have principally focused on in-vivo characterization of tumours.
One of its earliest applications was to analyse blood vessels in a brain tumour, since the blood-brain barrier (BBB) blocks the contrast agent in normal brain tissue, but not in vessels generated by a tumour.
The contrast agent’s concentration is measured as it passes between the blood vessels and the extracellular space of tissue, and then as it returns to the vessels. In tissues with healthy cells or high cell density, the re-entry of the contrast agent into vessels is quicker since it cannot pass cell membranes. In tissues which are damaged or have a lower cell density, the agent is present in the extracellular space for a longer duration.

Numerous DCE-MRI studies on the brain have researched the correlation between BBB disruption and diseases such as acute ischemic stroke, pneumococcal meningitis, brain metastases, multiple system atrophy, multiple sclerosis and Type-II diabetes. One of the most exciting areas of research is the difference in signal intensity profiles over time between Alzheimer’s disease patients and controls.

Tumours and DCE-MRI
Elsewhere, researchers have also established the benefits of DCE-MRI for differential diagnosis of tumours in the head and neck region, such as salivary gland tumours and lesions in the jaw bone. DCE-MRI has also been used to demonstrate the nature of a lymphoma and making a differential diagnosis versus other lesions.
Prostate cancer is becoming a major area of application for DCE-MRI. One of the key limitations to standards of care in the past was the need for random prostate biopsies after discovery of elevated PSA values. This often led to discovery of inconsequential tumours. Meanwhile, the very same biopsies sometimes missed out on significant disease. DCE-MRI, in conjunction with PSA, can identify tumours likely to cause death if left untreated.

Assessing response to chemotherapy
DCE-MRI is also being used to assess responses to chemotherapy. One example of an ongoing project in this area is CHERNAC (Characterizing Early Response to Neoadjuvant Chemotherapy with Quantitative Breast MRI), which is funded by the Breast Cancer Campaign in the UK.
Elsewhere, DCE-MRI has shown promise in detecting cancer recurrence. For example, biochemical relapse after radical prostatectomy can occur in as much as 15 to 30percent of prostate cancer patients. Detection of tumour recurrence in such cases can be difficult due to the presence of scar tissue. Determining the precise site of recurrence since patients with isolated recurrence could benefit from less-invasive treatments, such as radiation to the resection bed.
Other areas for DCE-MRI application include cardiac tissue viability – for example, to evaluate sub-clinical fibrosis and micro-vascular dysfunction. Researchers have also shown its utility in measuring renal function and partial/segmental liver function.

A full spectrum of methods
In general, the analysis of DCE-MRI is based on a full spectrum of methods from the qualitative to quantitative, with an intermediary semi-quantitative approach.

Qualitative analysis
Qualitative analysis is visual and depends on clinical experience and expertise. It assumed that tumour vessels are leaky and more readily enhance after IV contrast material is expressed. As a result, DCE-MRI patterns for malignant tumours show an early and rapid enhancement of the time-intensity curve (TIC) after injection of the agent, followed by a rapid decline. On the other hand, normal tissue shows a slower and steadily increasing signal after agent injection.

Quantitative analysis
Quantitative analysis is based on the pharmacokinetics of contrast agent exchange. It is complex, but allows for a degree of comparability. The limitation is due to a lack of standards. However, better and wider use of software has led to a growing consensus on approaches to quantitative analysis of DCE-MRI data.
One of the most widely used tools is the Toft and Kermode (TK) model, which is showing considerable promise in predicting and monitoring tumour response to therapy.

TK provides data about the influx forward volume transfer constant, KTrans, from plasma into the extravascular-extracellular space (EES). Ktrans is equal to the permeability surface area product per unit volume of tissue, and represents vascular permeability in a permeability-limited situation (high flow relative to permeability), or blood flow into tissue in a flow-limited situation (high permeability relative to flow). KTrans is known to be elevated in many cancers.

Pharmacokinetic modeling for analysing DCE-MRI dates to the early 1990s, and was followed by a consensus paper at the end of the decade ( Tofts P.S., Brix G., Buckley D.L., Evelhoch J.L., Henderson E., Knopp M.V. Contrast-enhanced T 1 -Weighted MRI of a diffusible tracer: Standardized quantities and symbols. Journal of Magnetic Resonance Imaging. 1999′).
Over the years, improvement of imaging techniques (e.g. higher temporal resolution and contrast-to-noise ratio) and greater knowledge of the underlying physiology have catalysed development of more complex pharmacokinetic models.
The TK model, for example, had been developed for measuring BBB (blood-brain barrier) permeability, and overlooked the contribution of the plasma to total tissue concentration. However, as the model gained popularity in assessing tumours throughout the body, vascular contributions to signal intensity were also included.

Semi-quantitative models
The semi-quantitative model seeks to fit a curve to data. Like the visual/qualitative, this approach also assumes early and intense enhancement and washout as a predictor of malignancy. However, semi-quantitative analysis also calculates a variety of dynamic curve parameters types after initial uptake, such as the shape of the time-intensity curve (TIC), the time of first contrast uptake, time to peak, maximum slope, peak enhancement, and wash-in and washout curve shapes.
Broadly speaking, there are three types of curve: Type 1 (persistent increase), Type 2 (plateau) and Type 3 (decline after initial upslope). One of the most attractive features of the semi-quantitative model is its relative simplicity in using parameters to differentiate malignant from pathologic but benign tissue.
For example, in the head-and-neck region, a rapid increase in TIC (fast wash-out pattern) indicates a strong possibility of Warthin’s tumour – a benign, sharply demarcated tumour. A persistent increase suggests the possibility of pleomorphic adenoma. A plateau pattern with a slow washout is characteristic of both a malignant tumour and adenoma.

In spite of enthusiasm about the semi-quantitative approach, it cannot be generalized across acquisition protocols and sequences as well as several other factors which impact on MR signal intensity. In turn, these affect curve metrics, such as maximum enhancement and washout percentage. Differences in temporal resolution and injection rates can also change the shape of wash-in/washout curves, making comparison difficult. Finally, such descriptive parameters provide no physiologic insights into the behaviour of the tumour vessels.

The limitations of DCE-MRI
DEC-MRI itself faces some major limitations. Firstly, there is a lack of standardization in DCE-MRI sequences and analysis methodology, making it difficult to compare published studies. In general, shorter acquisition times lend themselves to more comparability.
One frequent problem is movement by the patient and organ motion (e.g. in the gut, the kidney, bladder etc.). Since a DCE-MRI study procedure is over 5 minutes, there can be considerable misregistration between consecutive imaging slices, leading to noise in the wash-in and washout curves, and problems fitting pharmacokinetic models to the curve.
New DCE-MRI postprocessing software seeks to correct this by automatically repositioning sequential images for better alignment. However, these too do not use common algorithms to process the data and generate parametric maps and can show differences – e.g. in tumour vascularity. To enable further investigation of the value of DCE-MRI of the prostate, the technique of DCE-MRI and the pharmacokinetic model used to analyse it must become more standardized.

One of the most serious problems with DCE-MRI, however, is its non-specificity which can lead to to both false negatives and false positives.
Other sources of uncertainty in DCE-MRI studies include a lack of data. For example, one typical assumption is fast water exchange between compartments in spite of suspicions about the influence of restricted water exchange. Indeed, many quantitative models disregard intracellular space since it is assumed that there is no contrast media exchange. However, others have pointed out that water itself can exchange between the cell and the extracellular space, thereby influencing signal changes in the extracellular space. This is clearly an areas which calls for more study.
Further research is also required in areas such as relaxivity values for a contrast agent, field strength and tissue/pathology. Currently, relaxivity across tissues and compartments is generally assumed to be uniform.

To conclude, DCE-MRI is a significant and promising diagnostic modality. However, for most clinical applications, it cannot be used on a standalone basis, regardless of curve shape or intensity of enhancement. DCE-MRI needs to be viewed in the context of other MRI parameters such as diffusion-weighted MRI and MR spectroscopic imaging as well as T2-weighted MRI.

YGIA Polyclinic’s Radiology Department has been operating for 27 years and its staff is actively involved in ongoing clinical research and training to ensure the best possible services to all patients. A year ago, it embarked on setting up a breast imaging unit equipped with state-of-the-art technology, culminating in the installation of the Hologic Selenia Dimensions DBT system.  
Dr. Annie Papoutsou, Head of the X-ray Department gives us the full picture.

How has your hospital expanded in recent years?
In 2007, after major renovations and an extension of its building facilities, the YGIA Polyclinic private hospital boosted its capacity to 152 beds, extended the number of operating theatres to 12, established and extended the capabilities of its Clinical Laboratory Department, Radiology Department (X-ray, Mammog, Fluoroscopy, MRI, CT, Ultrasound), produced a multi-dynamic Intensive Care Unit (ICU), Obstetrics & Gynecology, and Pediatrics Departments. Furthermore, from mid June 2012, a state-of-the-art Cardio-Vascular Catheterizations Centre was established at the hospital offering the only 24-hour acute percutaneous coronary intervention (PCI) service in Cyprus. Moreover, the hospital has a range of fully equipped ambulances working 24 hours in order to be able to best respond to emergencies.

What type of equipment is used by your department?
Most of our X-ray rooms use the latest DR digital detectors providing superior quality images almost instantly, and are linked to an enterprise-wide fully integrated RiS/PACS.
Last year we organized a breast imaging unit, equipped with the latest technology, FFDM Hologic Selenia Dimensions, a GE ultrasound  with strain and shearwave elastography and a Hitachi ultrasound with high frequency  linear probe.
In our department we performed  more than 90,000 exams yearly.
What was the rationale for selecting the Hologic Selenia Dimensions system?
The decision was based on the special features offered by the Selenia Dimensions; these include:

• 2D imaging or combo mode (2D+3D) imaging in the same  compression.
• Exceptionally sharp  images with minimal dose.
• Streamlined workflow.
• Ergonomic design for comfort and  ease of operation.
• We believe that it offers the best technology available.

What are the advantages of Digital Breast Tomosynthesis?
The use of Digital Breast Tomosynthesis in breast screening enables us to find more invasive cancers than conventional 2D mammography alone.
The masses, distortions and asymmetric densities are better visualized with the Selenia Dimension.
But also it can reduce the costs associated with unnecessary recalls and it can reduce the incidence of negative biopsies.

Has the number of exams been affected by the adoption of DBT?      
Since the installation the number of mammography exams has increased up to 50%.
The recalls rate has decreased and also additional views have decreased.

When did you decide to acquire the C-View software and what are the benefits?
C-View software was installed from the first  day of the equipment installation, giving us the possibility of eliminating the need for conventional 2D exams after 6 months. The combined tomosynthesis and C-View exam makes lower patient radiation dose possible. Tomosynthesis exams with C-View software offer a patient dose similar to a 2D  only exam with  superior clinical performance for all breast types.

You are also using the Affirm biopsy device –what is the typical biopsy procedure followed in your department?
Up until now we are using the Affirm stereotactic biopsy device and in the next few days we are going to install the Affirm stereotactic tomosynthesis biopsy device to target lesions seen only with tomosynthesis. The typical biopsy procedures followed in our department are the biopsy under  the stereotactic  guidance and hookwire localization for subtle masses clusters of microcalcifications and architecture distortion.

What  do you see as the next step for improving the performance of your department?
Installing I-view with the contrast media.
I-View is a contrast-enhanced mammography technique that may be a viable alternative to breast MRI in performing contrast agent breast imaging. It offers certain advantages over MRI, including reduced cost and shorter procedure times. The imaging combination of contrast enhanced 2D imaging (CE2D) along with a 3D tomo scan, gives additional information beyond a CE2D examination alone, and may allow localization and morphologic evaluation of an enhancing lesion, further increasing the value of the CE2D procedure.

In spite of alarm bells that artificial intelligence (AI) would decimate the radiology profession, a host of barriers – both technical and regulatory – make this unlikely to happen for the foreseeable future. Instead, over the coming decade, AI is at best likely to help radiologists do their jobs more quickly and lead to improved patient outcomes.

From CAD to AI
AI in radiology, in some senses, has tended to raise the same level of expectation as computer-aided detection (CAD) did for the profession in the 1990s. Indeed, there is now a distinction between computer aided detection which reduces observational oversight and false negatives in interpreting medical images, and computer aided diagnosis (also called CAD) – by virtue of which software is used to analyse a radiographic finding to estimate the likelihood of a specific disease process (e.g. a benign versus malignant tumour).
As a result, in spite of tens of thousands of machine-learning algorithms, there is little connection to clinical application. Most remain confined to the realms of research.

The Black Box barrier
Radiologists, for example, use visual pattern matching. However, few object recognition algorithms have yet been tested on gray-scale images, such as those widely used in radiology.
Though specific algorithms could in principle be tailored for specific tasks, they use different assumptions and targets, and often are written to function in different modalities. Consolidating a set of algorithms into one package and then using this to underpin image or data analysis is not feasible.
In effect, the key problem with CAD detection is its black box’ nature, which means they cannot explain why an object has been identified as abnormal. Many users remain suspicious about sharing the already-grey zone between detection and diagnosis with a machine, which only provides probabilities.

Sensitivity and specificity
The above kind of issues also hinder AI. Nevertheless, the technology is rapidly evolving and may offer some solutions to new challenges.
Like radiologists, AI faces the twin pulls of sensitivity and specificity, between false positives which overcall disease and false negatives which undercall it. It is clear that it will favour sensitivity over specificity.

Technology creates its own momentum
In recent years, radiologists have been forced to cope with an explosion in the stock of medical images, thanks to modern imaging technologies and PACS storage capacity. In the UK, for example, almost 5 million CT scans are performed per year by the NHS. At the upper end, a single pan scan’ CT of a trauma patient, for example, renders about 4,000 images. Indeed, a busy radiologist can read about 20,000 studies a year.
To deal with this burden – both physical and visual – radiologists clearly need help. AI seems to have become one of the most optimal.
There is, nevertheless, some irony here. Technology, in this case consisting of new imaging modalities, has led to an increase in the workload on radiologists. This is in spite of the fact that the disease burden has remained more or less the same, as has the prevalence on imaging of clinically significant pathology. However, the growth of imaging stock has led to a sharp rise in the presence of detectable and potentially significant pathology. Radiologists therefore face the massive challenge of finding ways to use the latter. This is where yet another technology, AI, steps in.

Industry push combines with radiologist pull
While the need to handle the imaging data explosion will see radiologists pulling’ AI, industry has chosen radiology to push’ for clinical validation. There are two reasons for this: the sheer volume of the imaging data and its continuing growth make it a huge market, while the fact that it is stored in structured and computer-readable DICOM format means it is a ready one.

AI’s own dynamics in change
Meanwhile, AI itself has seen some changes. Although, fuelled by science fiction and Hollywood, the popular imagination associates AI with self-awareness, what we really still have is more accurately machine intelligence. The implications of even such a toned-down definition should, however, not be under-estimated. Neither should some recent developments.

From Deep Blue to AlphaGo
In the late 1990s, IBM’s Deep Blue supercomputer defeated grandmaster Garry Kasparov in a chess game. In March 2016, Google DeepMind’s AlphaGo defeated Lee Sedol, a 9th level Go grandmaster 4-1. For AI experts, the AlphaGo win is far more impressive than Deep Blue because Go is less rules-bound than chess.
Due to these constraints, Deep Blue analysed millions of potential combinations and outcomes, in what IT professionals call brute force’ calculation. No computer can yet achieve this with Go, which according to Business Insider’ (March 10, 2016) has ‘more than 300 times the number of plays as chess. Alongside continuous scenario analysis, top Go players require both experience and intuition’. This is why AlphaGo’s win was seen as a paradigm shift in AI.

Deep learning
Unlike Deep Blue’s brute force, AlphaGo used a programming method called deep learning’, with so-called neural networks, which are far more similar to human thought processes than traditional computing. Rather than seeking to map out every possible move combination, deep learning (DL) is a relatively-unregulated process by which a computer figures out why something is what it is, after being shown several examples. It uses a large but still-finite sample of data, draws conclusions from that sample, and then, along with some human inputs, repeat the process over and over again, to simulate millions of games into a decision-making system.
Technically, AlphaGo’s deep neural networks consisted of a 12-layer network of neuron-like connections with a policy network’ to select the next move and a value network’ to predict the winner of the game.

A new benchmark
Neural network-based deep learning is now the benchmark for AI in radiology, with IBM’s poster child Watson leading the way. At the 2015 RSNA meeting, Watson showed its capacity to find clots in brightly shining pulmonary arteries.
Watson, however, has a DL rival in Australia’s Enlitic, which has developed a lung nodule detector claimed to achieve positive predictive values that are 50percent higher than those of a radiologist. As the detection model analyses images, it learns from those images. It not only finds lung nodules, it also provides a probability score for malignancy. Enlitic is now conducting a trial on a model to detect fractures using X-ray images overlaid with a heat map to highlight their location within a conventional PACS viewer. The clinical application will eventually encompass X-ray, CT, and possibly MRI. At the moment, Enlitic is working to incorporate ACR guidelines into it.
Although both Watson and Enlitic use deep learning, the approach is different. Watson seeks to understand’ a disease, Enlitic simply seeks to find source problem data, solve it, and produce a diagnosis.

Another DL developer is MetaMind, since last year part of CRM (customer relationship management) giant Salesforce.com. MetaMind has an alliance with teleradiology provider vRad to identify key radiology elements associated with critical medical conditions, especially in the latter’s focus area of emergency departments (EDs). The first tool to emerge from the partnership was an algorithm to identify intracranial hemorrhage (ICH), often seen in ED patients and requiring prompt action. vRad, which has put the algorithm into a beta phase that will allow it to collect data to demonstrate outcomes, is adapting it to identify other critical conditions, such as pulmonary embolisms and aortic tears.

Swarm AI
Apart from deep learning, radiology is also seeing the first successful experiments with swarm AI, which helps form a diagnostic consensus by turning groups of human experts into super experts. The technology borrows from nature, which sees species accomplishing more by participating in a flock, school or colony (a swarm’) than they can individually. One study, published in Public Library of Science (PLOS)’, stated that swarm intelligence could improve mammography screening and has the potential to improve other types of medical decision-making, ‘including many areas of diagnostic imaging.’ Another study found that accuracy in distinguishing normal versus abnormal patients was significantly higher with swarm AI than the radiologists’ mean accuracy.

Challenges ahead
Nevertheless, there is much more to be achieved before AI becomes an everyday tool in radiology.
The biggest roadblock will consist of regulators, who are unlikely to sanction the use or marketing of intelligent’ machines. In the US, as first of their kind, they lack the predicate devices needed to be regulated under the FDA’s 510(k) rules, and it would take decades to get approval for each algorithm.
A second issue is the time and cost to get datasets to fine-tune the algorithms. Watson, for example, has a backlog of 30 billion medical images to review.
Thirdly, the algorithms would also raise significant legal and ethical issues, such as knowing when they could be trusted.
Finally, even were such machines to become available, referring physicians are unlikely to accept conclusions or interpretations drawn solely by them.
The scale of such challenges has already been seen by developers of computer-aided detection (CAD) algorithms – and the change of CAD to detection’ rather than diagnosis’, as it was called in the early days.

Need and benefit, reality checks
In short, for now, radiologists need AI just as much as AI needs them.
Radiologists will have to begin to work with AI, both to improve the technology itself and to reduce routine, repetitive tasks such as confirming line placements and looking at scans to find nodules.
On its part, AI is likely to become an increasingly smarter tool, to improve efficiency, for example by prioritizing cases, putting thresholds on data acquisition, improving workflow by escalating cases with critical findings to the worklist of a radiologist and providing automatic alerts to both radiologists and other concerned clinicians.
In the longer term, DL algorithms are likely to be trained to recognize disease patterns, identify, outline and measure nodules and possibly highlight suspicious areas in images. This is likely to be followed by the use of DL-based AI as clinical decision tools, for example to help referring physicians select or narrow choices of scans, based on clinical observations in an EMR. Such steps would not only free up resources for additional testing but also improve patient care, thereby making radiologists even more integral in the care management process.

In the final count, a resonant reality check on AI has been provided by Eliot Siegel, MD, professor of radiology at the University of Maryland. He has offered to wash the car of anyone who develops a program than can segment adrenal glands on a CT scan as reliably as a 7-year-old.