Quantitative Radiology

Applications to Oncology

Edward H. Herskovits1    Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland, Baltimore, Maryland, USA
1 Corresponding author: email address: ehh@ieee.org

1 Introduction

Oncologists, clinician-scientists, and basic scientists collect a plethora of data in the process of caring for patients, managing clinical trials, and investigating cancer biology. As we have developed more sophisticated means for noninvasively delineating and interrogating neoplasms, the resulting image data have come to play a central role in oncology. To understand the current impact and long-term promise of radiology with respect to oncology, it may help to characterize the nature of the information sought as we diagnose and treat cancer patients.

The ultimate goal of patient care in oncology is to maximize disease-free survival (DFS)—or, barring that, progression-free survival (PFS)—while minimizing the morbidity of treatment (i.e., to maximize quality-adjusted life years). Ignoring intercurrent illnesses and treatment morbidity for the sake of this discussion, we take PFS to be a function of tumor burden, which can be decomposed into two independent factors: the number of tumor cells and the malignant potential of each cell. For many years, the former—extent—was determined via exploratory surgery and summarized as tumor stage, and the latter—grade—was determined by pathologists from what was hoped to be a biologically representative sample obtained during this operation. Advances in radiology first became evident with respect to staging, for the simple reason that it is much easier to generate images that show macroscopic groups of cells than it is to generate images that show how these cells are likely to behave. Only in the last decade has radiology begun to offer information regarding tumor biology, and such information still pales in comparison with that obtained from histopathology and genetic analysis.

In parallel with the increasing complexity of image data, there has been steady progress in the quantification of these data. Although clinical radiology reports are unfortunately replete with verbiage such as “large mass in the right frontal lobe,” researchers have begun to deliver on the promise of computer-based methods for quantification of tumor extent and have also developed quantitative or semiquantitative methods for characterizing tumor biology. The premise underlying such efforts is that quantitative—rather than qualitative—indications of tumor extent and biology render more precise prediction of DFS, thereby promising superior patient care and assessment of therapy. Herein I explore the arc of radiology's contributions to oncology, both in terms of the information provided and efforts to quantify this information, with the expectation that such exploration will shed light on future developments in oncology research and practice.

2 Radiological Characterization of Tumors

The advent of computed tomography (CT) revolutionized the staging of solid tumors; since then, the quality and range of information provided to oncologists via noninvasive radiological examinations have steadily increased. The vast majority of this information relates to tumor extent; however, magnetic resonance (MR), positron emission tomography (PET), and newer modalities have offered progressively more detailed information about tumor physiology (Fass, 2008). Despite improvements in these modalities, there remain significant problems. For example, it is well known that, even with a combination of advanced MR sequences such as DTI and perfusion, we cannot accurately delineate the extent of infiltrative tumors, such as glioblastoma. Although there has been a striking expansion of research modalities for characterizing tumors (Budde & Frank, 2009; Cai & Chen, 2008; Desar et al., 2009; Fass, 2008; O'Connor et al., 2008; Pfannenberg et al., 2007; van der Meel et al., 2010; Weissleder & Pittet, 2008), we focus here on those most widely applied in clinical research and practice: CT, MR, and PET.

2.1 Computed tomography

The principal forms of CT used in oncology are structural (routine) CT, CT perfusion, and dual-energy CT (DECT).

2.1.1 Structural (routine) CT

CT, usually following intravenous iodinated-contrast administration, has been the workhorse of oncologists and researchers seeking to stage tumors and determine response to therapy. Relative to MR, PET, and other molecular imaging techniques, CT is inexpensive, fast, applicable throughout the body, and widely available, all of which are critical features of a modality that would be used to establish internationally accepted response criteria for a broad range of neoplasms. For many solid tumors—including some of the most common, such as lung cancer and gastrointestinal malignancies—the contrast between tumor and adjacent normal structures (i.e., tissue contrast) is sufficient to support delineation of lesions (i.e., to estimate stage or tumor burden). With the advent of helical CT (Van Hoe et al., 1997) and multidetector CT, spatial resolution (particularly in the z-axis) increased, allowing characterization of ever-smaller lesions.

2.1.2 CT perfusion

Although CT provides excellent anatomic information for most tumors, it provides little physiologic information about tumors. CT perfusion, in which intravenous contrast is administered as a bolus, and voxel-wise time-attenuation curves are computed from repeated scans (including a baseline noncontrast scan), is one of the most common means for obtaining information beyond precontrast or postcontrast attenuation values. The widespread availability, first of helical scanners, and subsequently of multidetector scanners, has promoted CT perfusion from a research tool into a commonly used clinical tool, with applications across organ systems and disease categories (Miles & Griffiths, 2003).

There are several categories of mathematical models that have been used to inform the calculation of perfusion parameters from the time-attenuation curve, with varying assumptions about the interactions among the contrast material (e.g., diffusibility; bolus contour), the patient's physical state (e.g., cardiac output), and tissue characteristics (e.g., collateral flow; differences between capillary and arterial hematocrit), among others. Virtually all perfusion analysis models ultimately invoke the Fick principle, which codifies conservation of mass—in this case, blood—in the perfusion model. The Fick principle models perfusion with a single (arterial) input that supplies a volume of tissue, which in turn drains into a single (venous) output. Under this model, all contrast must be either in an artery, perfusing tissue, or in a vein.

Two major groups of methods for perfusion analysis are those based on deconvolution, and everything else. Deconvolution methods are considered to be more accurate than alternative approaches, but are also more complex. Methods that do not employ deconvolution, such as the maximum-slope method, often rely on simplifying assumptions, such as the assumption that no venous outflow has occurred during the time interval of interest; although such assumptions are clearly not valid in most CT perfusion acquisitions, they may introduce only minimal parameter estimation errors. Deconvolution approaches assume that the concentration of the contrast agent in tissue is a linear function of flow to the tissue and the convolution of the arterial input function and tissue-specific characteristics (Nabavi et al., 1999); there are fewer simplifying assumptions than in nondeconvolution approaches. Mathematical deconvolution methods, such as singular value decomposition (Kudo et al., 2009; Ostergaard, Weisskoff, Chesler, Gyldensted, & Rosen, 1996), yield blood flow (BF), blood volume (BV), and other tissue-specific parameters. An additional parameter that is commonly employed in clinical practice is mean transit time (MTT), which is computed from flow and volume via the relation BF × MTT = BV. Perfusion is expressed as BF per 100 g of tissue, which can be computed from BF, BV, and a tissue-density conversion factor.

Although CT perfusion has been utilized primarily in evaluating stroke patients, it has also found an important role in the evaluation of neoplasms. To the extent that neovascularity reflects tumor grade, features derived from CT perfusion will prove prognostically useful, particularly...