samedan logo
home > ebr > autumn 2014 > made to measure
European Biopharmaceutical Review

Made to Measure

Medical imaging has provided doctors with the ability to see anatomically what could previously only be assessed symptomatically. In oncology, this has allowed doctors to detect, diagnose and evaluate tumours, and provide more informed decisions regarding patient care. Importantly, quantifi cation of tumour size has provided clinical trials with an objectively defi ned surrogate for the number of tumour cells in each lesion. Various criteria – all based on the change in tumour size – have been used to determine patient response to therapeutic intervention. In addition, World Health Organization (WHO) criteria has used bi-dimensional measurements to evaluate changes in total tumour load, based on an ellipsoidal or spheroidal tumour shape (1,2).

The Response Evaluation Criteria in Solid Tumors (RECIST) has simplifi ed cross-product measurements of all tumours with uni-dimensional measurements of a sample. Objective response using uni- or bi-dimensional measurements are also found in the Response Criteria in Malignant Lymphoma, and other criteria for various cancer types. These measurements enjoy widespread use over volume estimates because they do not rely on complex analysis algorithms, are easier for radiologists to use, and have the benefi t of historical and regulatory experience. However, many biopharmaceutical agents are not cytotoxic, thereby providing therapeutic responses that do not lead to a dramatic reduction in tumour diameters.

Assessing the entire tumour volume has greatly advanced the science of evaluating tumour cell propagation, and there are now sophisticated, validated and readily available methods that improve on the old criteria, which have relied on convenient estimates of tumour size, assuming a spheroid. Now, the use of a number of semi-supervised computational methodologies can greatly streamline the process of tumour volume measurement, which can be very laborious and subject to observer variability. Furthermore, examination of tumour morphology using dynamic growth models of the entire volume can estimate the effect of biopharmaceutical agents based on their mechanism of action, and not simply on cell death.

This article will review the benefits and costs of using volumes in novel biopharmaceutical oncology therapies. It will also address the ongoing efforts to introduce these methods as regulatory-acceptable endpoints within a controlled clinical trial.

Historical Assessment

Tumour burden is defined in a clinical trial by documenting all the lesions present on a baseline scan – typically computerised tomography (CT) or molecular resonance imaging (MRI) – choosing some to follow quantitatively, measuring them, and adding up the individual measurements to compare with subsequent visits for treatment effect. Historically, these measurements have used single longest diameter (LD) or bi-dimensional measurements as surrogates for lesion volume. While the WHO criteria requires measurement of all lesions, RECIST uses a representative sample of lesions, assuming that change in this sample represents a proportional change in the overall body tumour load.

After treatment begins, a decrease in tumour burden below a lower threshold is called ‘response’, and an increase above an upper threshold is called ‘progression’. Otherwise, the disease is characterised as ‘stable’. Such measures have generally been shown to correlate with disease course and clinical outcome. Regulators are familiar with these measurements as trial endpoints, and it is easy to train radiologists to use them. Representing the entirety of tumour morphology by LD is appropriate primarily for cytotoxic therapies that disrupt tumour growth by simply decreasing the number of viable tumour cells; however, this provides little information on the details of therapeutic effect. Most tumours are also not spheroidal, and linear measurements may not reliably represent tumour growth or response. In reality, LD measurements can prove insensitive to change.


In its simplest form, volume estimation is based on a spheroidal tumour, where there is a simple relationship between the LD and volume:

V = – πLD3

Rotated ellipsoids have a similar relationship that includes both LD and the shorter perpendicular diameter (SD):

V = – πLD * SD2

Where LD and SD are positively correlated to some degree, the total number of tumour cells for a homogeneous tumour volume (for example, no necrosis) is proportional to the volume. It can be shown that within some relatively minor constraints, the percentage change in volume is linearly proportional to the percentage change in LD. The WHO and RECIST criteria rely on this assumption, which is largely true for many small tumours where cells die homogeneously throughout.

However, estimation of volumes using these formulae is inaccurate for many tumours, and completely inadequate for complex shapes. Therefore, volumes have to be estimated by counting voxels within the tumour. The first step is segmentation – defining the boundary between tumour and surrounding tissue. All reliable, validated algorithms require observer supervision to correct segmentation errors. Both the algorithm and observer must be trained and calibrated to ensure that any bias is systematic, or that change in tumour volume is determined without bias. Once the tumour is segmented, it is possible to add the segmented volumes on each slice (planimetry) or use more advanced statistical methods, like k-Nearest Neighbor, semi-fuzzy c-means and random fields, to segment images that are rich in different types of information, such as data available from MRI.

Theoretically, volumes should be more variable than LD simply due to the added dimensions. However, for a given change in diameter, the effect on volume is dramatically larger. In the simple case of a spheroidal tumour, a 20% increase in the diameter is a 72% increase in volume, or a near doubling. This can be seen in the following equations:

V0 = Kπd03

V1 = Kπ(1.2d0)3

V1 = 1.72V0

Similarly, a 30% decrease in the diameter translates to a 66% reduction in volume. The differences in LD and volumetric changes can be even more dramatic when lesions have complex shapes, and LD measurements can end up being more subjective than objective. In these lesions, volume cannot be reliably estimated by unior bi-dimensional measurements, and must be explicitly measured. This property will be important when evaluating the dynamics of tumour growth.

There are challenges in the assessment of tumour size by volumetrics; volume estimation algorithms vary and may not be comparable. Completely manual segmentation, though generally accepted as the reference method, has the clear disadvantage of being labour-intensive.

Furthermore, a little-advertised but real limitation of volume measurements is that tumour boundaries should ideally be visible on every slice, whereas linear measurements need only one image with clear boundaries to provide a value. Most common volumetric segmentation methods allow for some inter-slice interpolation, and use computational methods to initially segment the tumour and then allow for observer editing. However, there are limits to the amount of missing slice data that can be accommodated.

Importantly, missing slice data will also affect the accuracy of a diameter measurement, and being able to measure one diameter does not guarantee that comparison across time-points is accurate. The Quantitative Imaging Biomarkers Alliance (QIBA) (3), a collaboration of academics, industry leaders and regulators, has published guidance on how to perform volumetric assessments with predicted precision estimates.

Growth Kinetics

Tumour growth kinetics is rapidly gaining popularity beyond an academic interest in biological modelling. It is a natural outgrowth of pharmacokinetic/pharmacodynamic modelling, but with longer timelines. The basic premise is that a decrease in tumour growth rate following treatment, compared to the period prior to treatment, is evidence of treatment-related inhibition of cancer cell proliferation. Each patient serves as an internal control, and multiple longitudinal measurements provide more information than change from nadir. Since other measurements, such as inflammation or vascularity, can be used as additional model components, tumour growth kinetic analysis seems ideally suited for biopharmaceuticals.

While growth kinetic modelling is possible using uni-dimensional measurements, non-spheroidal lesions and lesions whose growth is not uniform will be less sensitive to change, and may provide contradictory results, depending on the biopharmaceutical mechanism of action. Criteria using broad response categories, such as RECIST, only detect changes in tumour biology when the tumour changes size enough to be classified as a response or a progression. Many biopharmaceutical agents are cytostatic and their effect is to slow the growth of the tumour, rather than cause it to shrink. In theory, trials using RECIST would detect such an effect as a delay in the time to progression. However, a change in tumour growth rate can give evidence of treatment efficacy long before the effect would manifest in a trial measuring time to progression. Kinetic models can be as simple as a mixed model with repeated measures – where change in size before and after treatment are compared over at least three time-points – or as complex as a Gompertz function that is used to model non-linear, sigmoidal growth with resource competition:

Y(t) = Kexp(K’exp(αt))

The study design should consider the spacing of acquisition times to adequately estimate the growth parameter α. An important practical challenge is that this analysis requires imaging obtained before the patient begins study treatment, to establish a baseline growth rate. Pre-trial scans may be difficult to obtain when they were performed at other medical centres, and the quality of these scans cannot be assured.

Operational Issues

The advanced assessment techniques discussed require higher quality scans than older methods. Collecting consistently high-quality scans with the specified image acquisition parameters from the diverse sites involved in a large clinical trial can be a significant operational challenge. Volume algorithms that use statistical methods to mitigate the effect of image quality, as well as manual method interobserver variability, will be a driving factor in the near term for any use of volumes.

A central facility (such as an imaging core laboratory) can greatly facilitate this process by standardising, testing and monitoring performance. A good core lab employs trained radiology technologists who can assess each site’s equipment, technical capabilities and standard of care to make sure that the site itself can acquire the quality data the study needs. They conduct site training sessions on the protocol-specific scanning parameters and contrast requirements. During the course of a trial, these specialists inspect all scans and give sites rapid feedback when discrepancies are noted, minimising missing images.

One major consideration for volumes is the considerable time and expense of segmenting not just one tumour but, in the case of RECIST 1.1, five and possibly more. Well-trained analysts quickly become efficient in initially determining the boundaries of a tumour, greatly simplifying a radiologist over-read. In a sub-study, where trained technologists segmented over 10,000 lesion volumes, the following conclusions could be drawn on the process:

  • Some lesions defied any ability to provide reliable estimates of anything
  • Average time per lesion was around three minutes
  • Poor lesions for volumes were also poor lesions for LD
  • A break-in period of about 30 volumes (10 patients) was necessary to become efficient
Positive Offerings

Volumes provide a more complete information picture of the entire tumour morphology and can also use additional information to facilitate tumour growth kinetic modelling. Validated volume estimation algorithms are available since many have been successfully tested in the QIBA Challenge (4). When manual methods must be used, they do not appear to be especially burdensome when the observer is experienced on both the indication and the measurement system. However, training is essential to ensure quality.

Tumour growth kinetic modelling presents the promise of sophisticated estimation of the growth of the entire tumour – necessary to evaluate the effects of biopharmaceutical agents. Medical imaging has more to offer than simplistic measurement methods.


1. Miller A, Hoogstraten B, Staquet M and Winkler A, Reporting results of cancer treatment, Cancer 47: pp207-214, 1981
2. James K et al, Measuring response in solid tumors: Unidimensional versus bidimensional measurement, Journal of the National Cancer Institute 91: pp523-528, 1999
3. Visit: php?title=Main_Page
4. Visit: php?title=VolCT_-_Group_3A

Read full article from PDF >>

Rate this article You must be a member of the site to make a vote.  
Average rating:

There are no comments in regards to this article.

David Raunig is the Senior Vice President of Medical and Scientific Affairs - Imaging at ICON. He came to ICON after 16 years at Pfizer, where he was Director, Research Statistics. David plays a prominent role as a statistical expert in the development and qualification of biomarkers, imaging and other areas for use as endpoints in clinical trials, and serves as Chairperson for QIBA Technical Performance Metrology Working Group. He received his BS in Physics from the United States Naval Academy, an MS in Computational Mathematics from Rensselaer Polytechnic Institute, and a PhD in Biomedical Engineering from the University of Connecticut.

Gregory Goldmacher is Senior Director, Medical and Scientific Affairs, and Head of Oncology Imaging for ICON. He received his undergraduate education at the University of Chicago, and his MD and PhD in Neuroscience at the UT Southwestern Medical Center. Gregory did his residency training in radiology at the Baystate Medical Center, and was a fellow at the Massachusetts General Hospital and at Thomas Jefferson University, before joining ICON in 2010. He also holds leadership positions in QIBA.

Stephanie Owens
is Senior Director of Imaging Operations at ICON. She comes from a background as a Radiological Technologist performing CT and MRI in the clinical setting. For the past 11 years, Stephanie has performed a number of roles in the management and execution of clinical trials, and currently oversees the teams of technologists and radiologists at ICON responsible for collecting images from sites, ensuring image quality, preparing images for independent review, and carrying out the reviews. She has additional qualifications in business management from University College Dublin and Delaware Valley College.
David Raunig
Gregory Goldmacher
Stephanie Owens
Print this page
Send to a friend
Privacy statement
News and Press Releases

Moderna Announces FDA Advisory Committee Unanimously Votes in Support of Emergency Use for a Booster Dose of Moderna’s COVID-19 Vaccine in the U.S.

Moderna, Inc. (Nasdaq: MRNA), a biotechnology company pioneering messenger RNA (mRNA) therapeutics and vaccines, today confirmed that the U.S. Food and Drug Administration’s (FDA) Vaccines and Related Biological Products Advisory Committee (VRBPAC) recommended that the FDA grant an Emergency Use Authorization (EUA) for a booster dose of the Moderna COVID-19 vaccine.
More info >>

White Papers

Transforming Temperature Data Management Practices to Reduce Labor Costs and Improve Visibility and Control

Berlinger & Co AG

There are a number of temperature monitoring database options available to the biopharma supply chain professional and choosing the right one can significantly improve workflow which will reduce administrative time. This article will provide insight into practices for improving common processes through enhanced database functionality in the areas of managing data loggers by expiry date, storing and retrieving calibration certificates and acknowledging/documenting common temperature excursion events. Additionally, it describes the unique features and benefits of the Berlinger Smartview® data management system. Smartview® is a “Software as a Service” (SaaS) platform designed to efficiently and accurately manage both workflow and data in a regulatory compliant manner (21 CRF Part 11 & Annex 11). This comprehensive temperature data management system ensures reliability, accuracy, security, accessibility, and visibility to supply chain professionals, enhancing Good Distribution Practices (GDP) of temperature sensitive shipments.
More info >>




©2000-2011 Samedan Ltd.
Add to favourites

Print this page

Send to a friend
Privacy statement