A journey of discovery ... developing quantitative biomarkers for MRI of the brain

During the past decade, we have witnessed a paradigm shift in the use of radiology in medical practice. While radiology continues to provide a cornerstone of advanced diagnostics and quality care, increasingly, medical imaging is being used for follow-up and monitoring a large spectrum of medical conditions. Thanks to the development of new technologies, we have entered into an era in which imaging techniques contribute to our fundamental understanding of physiological processes in health and disease. In particular for disorders of the central nervous system, new imaging and post-processing techniques (perfusion imaging, diffusion weighted imaging, diffusion tensor imaging, diffusion kurtosis imaging, spectroscopy, etc), have made a significant impact on clinical patient management, therapeutic decision-making, and outcome prediction.

But, despite cutting-edge scientific developments and expanded clinical applications, radiologists still rely mainly on visual assessment (“eyeballing”) to interpret radiology examinations. The traditional process of “visual image interpretation” by radiologists has remained essentially unchanged for over a century. And while this approach may be more or less ‘good enough’ to suggest an initial diagnosis, it is definitely not acceptable for the interpretation of follow-up examinations. Radiologists encounter great difficulties when comparing current images with previous studies. It is extremely difficult to accurately assess subtle changes from one examination to the next (for example in the shape, size and structure of a tumour). This is due to variations in patient positioning, sequence, equipment, protocols and parameters, window settings, etc. In addition to these technical limitations, visual assessment is also prone to subjective interpretation variations, failures of perception, lack of knowledge and human error. These factors lead to significant inter-observer, and also intra-observer variations in the visual inspection of imaging data.

Fortunately, things are changing, due to the introduction of “imaging biomarkers” in research and, even more importantly, in daily clinical practice. The word “biomarker” implies a measurable parameter that can be used as an indicator of a particular disease or some other physiological state of an organism. In a “white paper”, the European Society of Radiology (ESR) stated that the development of new imaging biomarkers has a high impact in terms of patient management, assessing risk factors and disease prognosis. In particular, “imaging biomarkers” are of great value to extract quantitative, objective, reproducible parameters, thus improving the value of imaging in clinical practice.

The advent of imaging biomarkers to clinical neuroimaging is a game changer. They provide innovative ways to explore new research avenues, and approach clinical questions. Both the pharmaceutical industry and the regulatory bodies are increasingly relying on imaging studies to provide surrogate end points in clinical trials (a surrogate end point is defined by the National Institutes of Health as “a biomarker intended to substitute for a clinical endpoint”). It is important to find out what works, and what doesn’t, quickly, cheaply and efficiently. Quantitative imaging biomarkers are helping drug companies to make “go/no-go” decisions about new products; in many cases, this obviates the need for more expensive and time-consuming exploratory trials, and it saves time and money. Healthcare is progressing towards evidence-based and personalised medicine, and doctors are rapidly adopting decision-support tools, all of which require the input of quantitative data and objective metrics.

There is a strong need for objective biomarkers in clinical practice. Basically, any feature that can be detected on an imaging study can now be used to quantify specific biological processes. Examples of quantitative neuroimaging biomarkers include: volume measurements (hippocampus, gray matter, whole brain, etc), apparent diffusion coefficient (ADC), fractional anisotropy (FA) or mean diffusivity (MD), cerebral blood volume and flow (CBV & CBF), etc. Radiologists are adopting imaging biomarkers in combination with advanced image processing techniques. Table 1 shows an overview of clinically relevant biological parameters and the quantitative imaging tools to study them.

Multiple Sclerosis as a model for the rational use of MRI Biomarkers

To illustrate the growing importance of neuroimaging biomarkers, let us focus on patients with Multiple Sclerosis (MS). When doctors need to compare MRI examinations of the same patient, obtained at different time points, from different institutions, on different MRI machines, it is nearly impossible to accurately detect changes in the number of white matter lesions (‘lesion load’), or to identify and count new and enlarging lesions. It is like trying to count the black spots on a Dalmatian dog running at full speed. Moreover, in addition to changes in the number, shape and activity of demyelinating plaques, the brains of patients with MS also undergo subtle modifications in brain volume. These small changes are impossible to detect by visual inspection, and yet, they have important clinical consequences. Fortunately, today, for patients with MS, imaging biomarkers provide the neuroradiologist and neurologist with key information on how the disease is progressing and whether the patient is responding to the treatment. Several imaging biomarkers, such as volumetric assessment of brain structures (tissue segmentation), have been shown to have excellent sensitivity and specificity for diagnosis or prognosis of various neurological diseases.

A prerequisite for the adoption of neuroimaging biomarkers is standardisation of image acquisition, data processing and analysis and image interpretation (i.e. generating a report). MRI protocols and sequences must be reproducible, accurate and sensitive; quality assessment should be an integral part of this process. Methods used for analysis should be adequate and observer-independent. Ideally it should be possible to compare a biomarker in a single patient to a healthy control group (reference values). The key words in this chain of production are “standardisation” and “validation”. Individual radiologists need to rely on computers, which are able to handle large data sets, perform centralised analysis and automated quantification.

Neuroimaging biomarkers for measuring volumes and changes in volume

Two types of neuroimaging biomarkers can be distinguished: cross-sectional and longitudinal. In the cross-sectional approach, we extract and measure volumes in a 3-dimensional MRI dataset of a single subject. The volume of the whole brain, or part of the brain (grey matter, white matter, cerebrospinal fluid, hippocampus, …), can be computed through segmentation techniques. These methods rely on “segmenting” brain tissue from the surrounding scalp and other extracerebral tissues. The probabilistic modelling of voxel intensities exploits the fact that different tissue types have different MR image characteristics. Volumes in millilitres for each class can be computed, by simply multiplying the sum of the tissue segmentation over all voxels by the voxel volume. For example, in patients with MS, it becomes possible to perform volume measurements of, for example, total brain volume or FLAIR white matter hyperintensities.

Longitudinal neuroimaging biomarkers take into account two (or more) MRI scans of the same subject, obtained at different time points, to calculate volume changes in brain volume. This makes it possible to evaluate MS patients for progressive brain shrinkage (atrophy), a parameter reflecting neuro-axonal and myelin loss, and which is increasingly being used as an outcome measure in MS treatment trials. Longitudinal methods for brain atrophy typically match two MRI scans using registration techniques and directly extract small changes in brain volume from this process. A similar approach can be used for the longitudinal segmentation of white matter lesions.

Adding quantitative MRI biomarkers takes clinical neuroimaging to the next level. But in order to successfully introduce quantitative biomarkers in the clinical imaging pipeline, several important elements should be taken into account:

1. Accuracy and reproducibility: Many software applications offering cross-sectional brain volume measurements, based on a single MRI scan, have errors in the range of 1 – 1,5 %, which is too much if the expected yearly rate of volume loss is far less than 1%. In other words, measurement errors should be small enough to be reliable in individual patients, and the results should be clinically meaningful. Fortunately, there are now certain softwares such as MSmetrix (CE marked) or icoBrain (FDA cleared), which were developed especially for monitoring MS and provide measures for atrophy and lesion load with measurement errors as low as 0,13% for whole brain atrophy. It is only when measurement errors are so low that meaningful conclusions for individual patients can be drawn.

2.  Scan time is money. Currently, the average MRI protocol for MS patients represents about 20 to 35 minutes of scan time. Typically, such an optimal scanning protocol would include a 3D-FLAIR sequence, diffusion-weighted imaging, T2-weighted sequence, 3D-T1-weighted images before and after Gadolinium-chelate injection. Biomarkers will only be successful if they can be derived from the standard imaging protocol for MS, without the need for additional (lengthy) sequences.

3. Integration into the standard workflow: The idea is that biomarkers should help doctors, and not generate extra work. Radiologists and neurologists alike don’t have time to perform additional post-processing for every patient. Ideally, the post-processing should be linked to the patient informatics, to automatically generate reports that include biomarker information about the lesions and cerebral atrophy. The radiologist should report back to the neurologist, qualitatively if not quantitatively, about the lesion status and the atrophy of MS patients, covering the following points: comparison with previous scan(s); evidence of new disease activity; number of new lesions (T2/T1); lesion size; overall assessment, including presence (definite/probable) and extent (number of new/enlarging lesions or gadolinium-enhancing lesions) of disease activity; change in T2 lesion volume; and evidence of brain atrophy.

How to transmit this information to the clinician

Today, MRI biomarkers are already an important factor for making therapeutic decisions. Efficient patient follow-up requires effective and consistent communication between the neurologist and radiologist. Unfortunately, most MRI reports are still written in prose, and do not make use of the full potential embedded within the MRI data sets. I strongly believe that communication regarding MRI findings between the (neuro)radiologist and the neurologist can be improved with automatically computed, quantitative values for the relevant imaging biomarkers. To this end, the (neuro)radiologist should have easy access to approved tools for calculating these biomarkers. Furthermore, when following the evolution of changes in an individual patient, comparisons could be made of biomarker values against relevant populations (e.g., healthy controls, MS patients that respond well to therapy, etc). Obviously, relevant confounding factors (such as age and sex) should be taken into account.

Conclusion

The introduction of (neuro)imaging biomarkers has led to a significant improvement in the diagnosis, management and follow-up of patients with MS. Standardisation of MRI acquisition protocols, and improvement of quantitative reporting tools provides a better understanding of the natural history of MS, and allows accurate treatment monitoring, for the greater benefit of patients.