Advances in
medical imaging analysis shedding new light on disease structure and
function over time. In this article, we examine the potential of imaging
biomarkers and other studies to improve the drug discovery and development
process at every stage.
| Sep
1, 2005 |
| By:
Edward
Ashton |
| Pharmaceutical
Discovery |
|
In the past decade, medical science has
made great strides toward the understanding of cellular biochemistry and
the mechanisms of disease. But as basic science has blossomed with the
sequencing of the genome and the advent of proteomics, the applied
sciences have failed to keep pace. Despite tremendous innovation in the
methods used to derive new compounds, the prevailing methods for testing
them have remained essentially the same for decades. A disease state is
induced in a cohort of laboratory animals — tumor cells are injected
under the skin, for example — and later the experimental treatment is
administered to some or all of the subjects. After a period of weeks or
months, the animals are sacrificed and a pathologist makes an assessment
of disease progression or regression.
This histology provides a glimpse of the
disease state at one point in time. But what did it look like before
treatment was started? What would it have looked like in another six
weeks? Did the experimental compound behave in the way that scientists
expected it would? That the prevailing method of drug testing in
preclinical studies fails to answer these questions and others underscores
a fundamental imbalance in drug discovery today that was addressed in the
comprehensive FDA white paper, "Innovation or Stagnation, Challenge
and Opportunity on the Critical Path to new Medical Products" (1).
The paper repeatedly made the point that the drug industry is using the
tools of the last century to evaluate this century's advances.
The development of this white paper, under
the leadership of former FDA commissioner Mark B. McClellan, marked a
seminal event in the history of drug development in the United States. In
the document, FDA calls for new development tools, specifically mentioning
imaging as a potential solution to the drug development bottleneck faced
by the pharmaceutical industry. Rarely has there been such early goal
alignment between FDA and the pharmaceutical R&D community around new
approaches to drug development.
A new field — let's call it imageonomics
— rapidly is emerging. Just as the sequencing of the genome led to
genomics and the study of the role of proteins led to proteomics, so
advances in medical imaging analysis technology are creating a field that
sheds new light on disease progression by enabling the precise measurement
of small changes in structure and function over time. This opens a new
quantitative window into disease progression. The measurement units of
this field, image-based biomarkers, have been singled out by FDA as
promising tools for every stage of drug development, from preclinical
research to the approval process.
At the Fifth National Forum on Biomedical
Imaging in Oncology (Bethesda, Maryland, USA) on January 29, 2004, the
director of FDA's Center for Drug Evaluation & Research, Janet
Woodcock, said, "There is tremendous potential for the use of imaging
in drug development...from preclinical [applications] all the way to using
surrogate markers for approval." While the use of imaging endpoints
by FDA for drug approval has been somewhat limited in the past, the white
paper and subsequent comments signal a readiness to assess their use in
the future.
Not surprisingly, leading pharmaceutical
companies such as Pfizer (New York, New York, USA) are discovering the
enormous potential for medical imaging in drug discovery and development
to close the gap between the burgeoning basic sciences and the lagging
applied sciences. These companies have invested millions of dollars in
developing medical imaging technology.
While it generally is acknowledged that
medical imaging holds the potential to make drug discovery and development
more efficient, the best practices for applying it to preclinical and
clinical research still are emerging. At the preclinical stage, advanced
medical imaging analysis provides a powerful tool to assess method of
action, ensure the transition of data from preclinical to clinical studies
and achieve the reproducibility required for studies that use a relatively
small number of subjects. In this article we will discuss the practical
aspects of applying quantitative medical imaging to preclinical research
and the substantial savings in time and money that imaging lends to drug
discovery and development.
Room for
Improvement
Recent
scientific advances including the development of targeted therapies and
the sequencing of the human genome have brought about massive changes in
the business of drug discovery and development. Large companies that once
had relatively few compounds in development now juggle hundreds of
potential drugs in the pipeline. For the industry to sustain itself,
companies must devise ways to quickly identify promising compounds and
cull ineffective ones, because the price of failure is enormous. Acting
FDA commissioner Lester Crawford stated that in 2002 some estimates put
the average cost of developing a new drug at $1.7 billion (2). With this
level of spending, says Crawford, "A mere two percent improvement in
predicting product failures in clinical trials could save $100 million in
development costs per drug." The Tufts Center for the Study of Drug
Development estimated that it costs $808 million, on average, to develop
and win market approval for a new drug in the United States (3). The study
states that better preclinical screens, to increase success rates from the
current 21.5% to one in three, could reduce capitalized total cost per
approved drug by $242 million.
The
Importance of Automation
Medical
image analysis can reduce late-stage attrition dramatically because it
offers more accurate information much earlier in the drug development
process. As an example, dynamic contrast-enhanced magnetic resonance
imaging (DCE-MRI) is able to provide information about blood flow and
vessel permeability in tumors. This information allows a relatively small
trial to quickly determine whether a compound designed to attack the
tumor's blood supply is having the desired effect. Whereas Phase I
clinical trials normally focus on dosage and safety, studies that
incorporate imaging analytics also can test for drug efficacy, offering
information that can save millions of dollars by allowing companies to
better prioritize their drug pipelines and make go/no-go decisions much
earlier. In preclinical research, scientists can test a method of action
and lay the foundation for a more streamlined clinical trial process from
Phase I through Phase III by obtaining critical information about
efficacy.
Figure 1. An MRI scan of the brain
of a multiple sclerosis patient. Notice that the lesions are
irregular in shape, with indistinct borders. This makes precise
specification of the lesion boundaries very difficult to do by
hand, leading to the potential for high variability among
analysts.
|
The value of
medical image analysis stems from quantification and automation. While the
primary shortcoming of standard endpoints (such as pain or functionality
scoring) is that they are largely subjective and difficult to reproduce,
imaging allows the replacement of a subjective evaluation — knee pain
ranked on a scale of 1 to 10 — with an objective quantification —
cartilage volume in cubic millimeters. Two subjects with the same degree
of disease might report very different levels of pain. Quantitative
imaging removes this arbitrary aspect from the clinical trial process (4).
Automation in the image analysis process — using a computer algorithm to
measure lesion size rather than a clinician with a ruler — provides a
degree of accuracy and reproducibility that cannot be duplicated by manual
techniques. A good example of this phenomenon is provided by the
measurement, using magnetic resonance imaging of lesion burden in multiple
sclerosis (MS) patients. MS lesions generally are irregularly shaped and
tend to have fuzzy, indistinct boundaries (Figure 1). As a result, if two
well-trained radiologists are asked to measure the total lesion burden for
a given MS patient by manually tracing the lesion boundaries, the results
they generate will differ by an average of 20%. Even asking a single
radiologist to analyze the same data set twice likely will result in
measurement variability of 7% or more. Introducing automation into this
process can reduce this variation to 2% or less (5), allowing clinical
trials to be carried out with a fraction of the subjects and in
significantly less time.
For
preclinical studies, reproducibility is equally critical because
preclinical groups generally have more limited budgets and timelines,
coupled with the need to evaluate increasing numbers of experimental
compounds. Highly reproducible measurement techniques allow these groups
to achieve statistically significant results with a relatively small
number of subjects.
Precise,
automated measurement brings another critical benefit: it enables the
detection of small changes in structure and function over time. In
evaluation of osteoarthritis, for example, MRI of the cartilage in the
knee coupled with automated measurement of volume and chemical composition
shows disease changes in months; these changes would not be apparent using
standard X-ray evaluation for years. With this quality of information,
researchers can more confidently make the go/no-go decision for a compound
early in the evaluation process, allowing scarce resources to be allocated
to the most promising candidates.
Reproducible
medical image analysis is driven by algorithms that enable quantitative,
volumetric measurement of structures and metabolic functions. Guided by
the information present in the images, as well as embedded anatomical
knowledge, the algorithms enable segmentation of different tissue types,
such as bone, muscle, fat and fluid. From an MRI knee scan, for instance,
it is possible to produce a three-dimensional (3-D)reconstruction that
graphically distinguishes cartilage from underlying bone, as well as from
ligaments, fluid, degenerated menisci or inflamed synovium. This
capability provides a valuable assessment tool for clinical research in
osteoarthritis — a disease with multiple endpoints — because it allows
the very sensitive and specific measurement of all the components of the
knee joint and the detection of small changes in any of those components
over time.
In
oncology studies, the benefits of a volumetric, automated approach to
image analysis become abundantly clear. The current standard approach to
tumor measurement is based upon the response evaluation criteria in solid
tumors (RECIST) guidelines. RECIST measurements are uni-dimensional
(longest diameter) and limited to the axial imaging plane. These
measurements often are made from printed films using a ruler or a pair of
calipers.
Figure 2. A tumor pre-treatment (a)
and post-treatment (b). Standard uni-dimensional measurement had
shown only a 5% decline in size. However, volume analysis shows
this tumor has lost over 70% of its bulk.
|
Most
researchers will concede that a single diameter measurement does not
provide all the information needed to determine if a tumor is responding
to therapy in many cases (Figure 2). However, a full evaluation of the
tumor volume generally has been considered to be too time-consuming and
not reproducible enough for use in clinical trials. Automated tumor
segmentation addresses both of these issues, providing accurate and
reproducible volumetric measurement without requiring a great deal of
valuable time from oncologists or radiologists.
Functional
Imaging Techniques
In
addition to structural measurements such as size, thickness or shape,
properly utilized medical imaging can allow the assessment of functional
parameters. Functional imaging encompasses a wide variety of imaging
techniques, including functional MRI (fMRI), DCE-MRI, dynamic
contrast-enhanced computed tomography (DCE-CT) and positron emission
tomography (PET). These methods allow the assessment of the metabolic
activity of an organ or lesion through the measurement of markers such as
tissue blood volume, blood flow, oxygen utilization or glucose metabolism.
In the
clinic, functional imaging makes it possible to distinguish between scar
tissue and viable tumor, and in some cases between benign lesions and
malignant ones. In drug development, functional imaging allows the direct
measurement of drug effects that otherwise would only be observable
indirectly, through their influence on patient survival or well-being. For
instance, functional measurement allows researchers to test the efficacy
of anti-angiogenesis compounds that cut off blood flow to tumors. Several
functional imaging modalities — including DCE-MRI, DCE-CT and dynamic
PET — allow the direct measurement of parameters such as blood flow,
blood volume, and vessel permeability. These methods work by measuring the
concentration of an injected tracer in the blood and in the target tumor
every few seconds for several minutes. The form and magnitude of the
resulting time-concentration curves then are fitted to a model that
produces the measured parameters (6). However, functional imaging does
impose an additional burden on the radiologist. Obtaining a measurement of
tissue blood flow from a DCE-MRI scan requires the application of a
complex statistical model, and currently there is no standardized
commercial software available to accomplish this. Various academic groups
have shown coefficients of variation for DCE-MRI measurements ranging from
6% to 45%. Clearly, it is vital for researchers and clinicians interested
in these modalities to partner with individuals or companies with
experience and expertise in both image acquisition and data modeling and
measurement.
Measuring
vascular perfusion and permeability using DCE-MRI has proven to be an
effective tool for determining the efficacy of anti-angiogenesis compounds
that are intended to cut off blood flow to tumors. Pfizer recently used
DCE-MRI in a clinical trial setting to evaluate the effects of an anti-angiogenesis
compound designed to inhibit the action of multiple targets, including
vascular endothelial growth factor (VEGF) and tyrosine kinase (TK) (4).
The primary objectives for the Phase I study were to determine the maximum
tolerated dose and to assess safety. In the past, these would have been
the study's only goals. However, because Pfizer now recognizes the
imperative to kill ineffective compounds as quickly as possible, DCE-MRI
also was built into this trial to determine whether the compound was
having the desired inhibitory effect on vascular function.
Because
early-phase trials make use of a relatively small number of subjects,
measurement precision is vital if any useful information is to be taken
from them. For this reason, Pfizer contracted with VirtualScopics
(Rochester, New York, USA) to provide supervision and quality assurance at
the imaging sites and high-quality functional analytics. This trial was
able to successfully demonstrate a strong relationship between drug dosage
and vascular effect measured by DCE-MRI and also some correlation between
vascular effect and tumor shrinkage. Because of this trial, Pfizer was
able to confidently push this compound further down the development
pipeline (4).
Applications
in Neurology and Cardiology
In
evaluating diseases of the brain, automated medical image analysis is
particularly useful. Because the brain has no moving parts and has a
fairly consistent structure from person to person, it is possible to
generate a generalized map, or anatomic atlas, of the location and shape
of many of its important structures. Under the supervision of a
neuroradiologist, this map can then be applied to a series of patient
scans to provide a consistent measurement of neural structures that
frequently have unclear or even arbitrary boundaries.
Figure 3. The hippocampus, as seen
in two MRI brain scans. (a) The boundary between the hippocampal
head and the amygdala. (b) The boundary between the hippocampal
tail and the caudate nucleus.
|
A good
example of an important but difficult to measure neural structure is the
hippocampus, a gray matter structure of the brain that is involved in a
number of functions, including the formation of long-term memory. Changes
in the hippocampus are implicated in a number of diseases, including
intractable temporal lobe epilepsy and Alzheimer's disease. The
hippocampus is adjacent to and difficult to separate from other gray
matter structures, including the amygdala and the caudate nucleus (Figure
3). Manual measurements of the hippocampus are difficult to reproduce
because there is no clearly visible boundary with these structures. An
automated measurement technique using an anatomical atlas might not always
agree with any particular radiologist. However, experiments have shown
that any given radiologist is unlikely to precisely agree even with
himself if asked to measure the same scan a week or two later (5, 7). When
tracking changes over time the key factor is reproducibility, and in that
area automated methods provide a significant advantage.
Medical
image analytics also have shown great promise in cardiology and
angiography. It was long thought that the key danger sign in the
assessment of arterial disease was vascular occlusion — the narrowing or
blockage of the coronary or carotid arteries. However, it is now known
that large vascular plaques can form in key arteries without narrowing the
lumen at all, by pushing the outer wall of the vessel outward rather than
pushing the inner wall inward. Such plaques either can be relatively
harmless or deadly, depending on what is inside them and how likely they
are to break open, spilling their contents into the blood stream. The most
dangerous plaques have thin walls and large liquid cores filled with
lipids and other substances. When these plaques burst, the contents
quickly form clots, which can then lodge in the brain, heart or lungs.
With the
proper acquisition parameters, MRI can allow the identification and
measurement of vascular plaques. More importantly, it also can allow the
determination of the plaque composition. This makes it possible for
surgeons to distinguish between a relatively benign plaque that can be
left for future observation, and a potentially deadly one that requires
immediate surgery.
Assessing
Method of Action
In
preclinical studies, perhaps the greatest value of imaging analytics lies
in early assessment of method of action. Whereas histology provides
limited insight into the effects of a compound, image analysis
demonstrates whether or not the drug behaves in the way that researchers
predicted it would. Did a compound bind to a particular receptor? Did it
inhibit blood flow or the development of blood vessels? Did it attack
metabolic activity? Having these questions answered at the preclinical
stage enables researchers to optimize clinical research from Phase I
through Phase III.
Biomarkers
Medical
imaging analysis enables researchers to take advantage of recent
innovation in the development of image-based biomarkers from the nation's
universities and from organizations such as the Center for Biomarkers in
Imaging (Charlestown, MA, USA), a joint effort between Massachusetts
General Hospital (Boston, Massachusetts, USA) and the Harvard-MIT Division
of Health Sciences and Technology (Cambridge, Massachusetts, USA). The
introduction of new biomarkers provides greater opportunity to study the
mechanisms of action in a preclinical setting. One advantage here is that
researchers' predictions can be verified through histology at the end of
the study.
It is
important to bear in mind, however, that these biomarkers are not easy
either to acquire or to extract and analyze. Full exploitation of the
information available through medical imaging requires access to a
multidisciplinary team, including radiologists, physicists and software
developers. It is tempting for research groups with strong expertise in
one or two of these areas to believe that they can avoid the need for the
others. Academic groups in engineering, for example, are well known for
developing answers to questions that the medical community is not asking,
while radiology groups commonly attempt complex analytics using
undocumented freeware. Tight communication between these groups is
absolutely essential for the successful application of image-based
biomarkers to either academic or clinical research.
Translational
Medicine
The
translation of preclinical results to clinical trials traditionally has
been one of the most challenging stages in drug development. One key
reason has been the inability to follow a consistent set of biomarkers
from preclinical studies, where the primary endpoint might be histology
after dissection, through to Phase III human trials, where the primary
endpoint might be a subjective pain score. Imageonomics bridges this gap
by allowing a single set of biomarkers — for example, cartilage volume
assessed by MRI — to be applied both to preclinical and clinical
studies. This greatly increases the chances that results in human trials
will show some consistency with those in preclinical animal experiments.
The key
to successfully transitioning preclinical work into clinical studies lies
in minimizing the differences between the two. The imaging protocols used
in the preclinical and clinical studies should be as consistent as is
practically possible. In a DCE-MRI imaging study, for example, the
injection processes for the contrast agent should be comparable, as should
the frequency of imaging. At the same time, the analytics must be
consistent. If possible, the analysis software that is intended for use in
the clinical studies should be used for the preclinical work as well. This
means that preclinical studies must be carried out using
regulatory-compliant software and processes, which adds an additional
burden to the preclinical research group. However, the benefits both in
terms of consistency with clinical research and improved reliability for
the preclinical analytics themselves generally will outweigh the
additional costs.
Applying
Medical Imaging to Pre-clinical Research
To
incorporate medical imaging into a study, two major decisions first must
be made: 1) how to acquire the images and 2) how to analyze the images
once they have been acquired. As the use of medical imaging in preclinical
research still is emerging, there is no established practice ready to
handle image acquisition. Companies either outsource to facilities that
offer imaging equipment and technicians to acquire the images for them, or
pay for the use of a private facility and handle image acquisition
themselves. For image analysis, companies can go one of two ways: they can
hire the expertise internally, or outsource it to specialists who have the
resources already in place. Automated analysis requires software experts
to develop analysis algorithms and graphical user interface engineers, as
well as provide staff to handle the database reporting tabulation and to
conduct the analysis. Because of the breadth of expertise required, even
large companies such as Pfizer, which boast in-house imaging centers, have
opted to outsource to companies that specialize in medical image analysis.
Conclusion
Through
precise quantitative measurement, imageonomics offers a solution for the
imbalance that now exists between the basic and applied sciences in drug
discovery and development. Medical image analysis stands to greatly
improve the efficiency of preclinical and clinical tests for drug
efficacy. It also promises to reveal new and critical insight into the
mechanics of disease progression that will benefit all of medical science.
The move toward the precision of automated, quantitative measurement
represents a natural progression in an industry where accurate
information, determined earlier in the process, can make a difference of
hundreds of millions of dollars in the costs of developing a single drug.
This emerging field of imageonomics offers new solutions to solving the
drug bottleneck and overcoming the enormous cost of late-stage attrition.
Edward
Ashton is vice president of product development at VirtualScopics.
Edward Ashton can be reached at VirtualScopics, 350 Linden Oaks,
Rochester, New York 14625 USA. Tel. 585-249-6231 ext. 212; e-mail ashton@virtualscopics.com
*To whom
all correspondence should be addressed.
References
1. FDA
Web site. http://www.fda.gov.
Accessed March 2004.
2. L.C.
Crawford, Banc of America Securities Healthcare Institutional Investor
Conference, July 7, 2004.
3. Tufts
Center for the Study of Drug Development, Impact Report, Volume 4, Number
5, September/October 2002.
4. T.
McShane et al., Proc. ISMRM 11, 154 (2004).
5. E.A.
Ashton, C. Takahashi, M.J. Berg et al., J. Magn. Reson. Imaging 17,
300–308 (2003).
6. P.S.
Tofts, J. Magn. Reson. Imaging 7, 91–101 (1997).
7. E.A.
Ashton, K.J. Parker, M.J. Berg and C.W. Chen, IEEE-Trans Med Imaging 16,
365–371 (1997).
|
