What is a qualitative imaging? The term “quantitative imaging” may seem like an oxymoron to some or at least an unattainable goal. Neuroradiological practice often entails making a collective assessment of a set of qualitative “weighted” MR images and providing a concise narrative explaining their clinical significance. However, T1- T2-, D-, … weighed images are susceptible to a myriad of experimental confounds and artifacts, and often cannot be quantitatively compared in longitudinal and particularly, in multi-site studies.

What is the goal of quantitative imaging? A goal of quantitative MRI is to transform the radiologist’s MRI scanner into a scientific instrument that produces robust, accurate, precise, and stable measurements of quantities or parameters having a physical meaning, and hopefully a concomitant biological significance that can improve diagnostic sensitivity and/or specificity, enabling a more informed clinical assessment.

What characteristics and properties are shared by quantitative imaging metrics? Quantitative imaging parameters should share several common characteristics or features. They should either be intrinsic physical quantities or constants, which are defined in terms of a well-accepted physical or chemical model; or parameters, which represent a meaningful ratio of physical quantities or constants that measure an intrinsic feature or quality of a physical process. Image-derived physical quantities should be measured in physical units, robustly, accurately and precisely. One should be able to relate these quantities to a measured MR signal using well-established physical-mathematical models. It is prudent for some scalar parameters to be unitless or dimensionless. For directional processes, like anisotropic diffusion, scalar parameters should be rotationally invariant (1, 2), rendering them independent of the choice of the laboratory frame of reference or the position and orientation of the subject or subject’s tissue being scanned.

What are examples of quantitative metrics? Common quantitative metrics in diffusion MRI applications include the Trace of the apparent diffusion tensor (1, 3) or one-third of it -- the mean apparent diffusion coefficient (mADC), which characterize the isotropic part of the tissue water diffusion process. Dimensionless scalar parameters that measure the degree of diffusion anisotropy, like the Fractional Anisotropy (FA) (4, 5) (or the Relative Anisotropy) represent ratios of different features of the anisotropic diffusion process. It is critical that such scalar parameters, like the FA, are also designed to be rotationally invariant so that they reveal intrinsic features of the underlying physical process independent of the experimental design, details and laboratory or subject coordinate system.

What is the value of quantitative metrics? From a bench, preclinical, and clinical perspective, our ability to more sensitively and selectively extract useful microstructural features, independent of experimental design, hardware used, … should enable us to better follow changes in normal and abnormal development, disease progression, tissue degeneration, sequelae of brain trauma and effects of normal and pathological aging processes. In an era of precision medicine, they should improve our ability to make genotype-phenotype correlations, as in the ENIGMA DTI project (6).

Challenges and requirements of measuring and mapping quantitative imaging metrics: Generally, “It takes a pipeline” to measure and map a quantitative imaging metric. Quantitative imaging is significantly more challenging than weighted imaging, and much more costly and labor intensive. One must have a robust experimental design to enable the measurement robustly, accurately and precisely. There must be a means to calibrate the scanner for a measurement to assess artifacts, image quality, etc. There needs to be an accounting of all known or foreseeable experimental, biological, and imaging artifacts and a means to correct or at least mitigate their effects. There should be a quantitative assessment of the effect of the image reconstruction process on the generation of the MR signal of interest. There has to be a physico-mathematical framework in place to estimate quantities or parameters from the imaging data. There has to be framework for mapping and analyzing the data using a rigorous statistical approach, such as hypothesis testing. There should be methods in place for segmenting and clustering imaging, and a means to do meaningful group (population) analysis.

What are candidate higher-order or “advanced” quantitative imaging parameters or metrics? In the field of diffusion MRI, the advent of higher-order tensor (HOT) models of diffusion, such as Generalized Diffusion Tensor Imaging (7, 8) Diffusion Kurtosis Imaging (7, 9) and Mean Apparent Propagator (MAP)-MRI (10) have enlarged the family of potential quantitative imaging quantities and parameters to characterize features of diffusion processes occurring in tissue that do not necessarily obey the classical Einstein equation. These models can be viewed as “non-parametric” because they do not explicitly contain material or microstructural parameters to estimate or measure. MAP-MRI is an implementation of Callaghan et al.’s “k-space and q-space imaging” (11) that requires a vastly reduced amount of DWI data to construct the “average propagator” or MAP. Several new candidate quantitative metrics can be derived from extracting features of the size, shape, and orientation of these MAPs in each voxel. One such feature is the Non-Gaussianity (NG). This represents the fraction of the MAP that cannot be fit by a DTI or Gaussian diffusion model. Another is the Propagator Anisotropy (PA). This quantity reports the fraction of the propagator that cannot be described by an isotropic displacement distribution. In some ways, this is a generalization of the FA in DTI. Another quantitative parameter of interest is the return-to-origin probability (RTOP). In a restricted diffusion process, RTOP can be shown (in the limit of long diffusion times) to be inversely related to the volume of a confining pore. RTOP’s dependence on diffusion time may make it less robust as an imaging biomarker than other MAP-MRI derived parameters

What candidate higher-order or “advanced” imaging parameters or metrics arise from parametric modeling approaches? There have been several parametric models of diffusion within a voxel presented in the past decade. Parametric models, such as NODDI (12), CHARMED (13), ActiveAx (14), and AxCaliber (15), all ascribe the total MR signal to a sum of contributions from different distinct, non-exchanging compartments, with different assumed morphologies. The features that these models yield, such as the intra-axonal fraction, neurite density, extracellular fraction, … are potentially useful in characterizing tissue microstructure and their possible alterations in pathology.

Overhanging issues with proposed quantitative parameters from “advanced” imaging methods? For non-parametric models like DKI, one must ask about the possible robustness of kurtosis tensor-derived parameters, including the mean Kurtosis. These parameters are measured from DWIs using a model that is quadratic as opposed to a linear with respect to “b”. Diffusion Kurtosis Imaging (DKI) parameters are therefore sensitive to the range of b-values used and will invariably will change if higher order terms are included in the diffusion model. DKI parameters also inherently sensitive to the timing parameters of the diffusion gradients (i.e., the pulse duration and diffusion time), used to acquire the DWIs. For parametric modeling approaches, one must inquire about whether the proposed assumed microstructure is consistent with the ground biological truth. More testing and vetting are required to assess these parametric models before parameters or features derived from them are promulgated from bench to bedside as quantitative metrics or biomarkers.

Other candidate quantitative parameters: Another class of potential quantitative parameters could be based on the hierarchical character of many biological tissues. Structural tissues, such as tendon, ligaments, and skeletal muscle are characterized by a “bundles-within-bundles” fractal-like architecture. This architectural paradigm is also present in white matter pathways in the CNS and nerve fascicles in the PNS. Extracellular matrix and brain parenchyma also have a hierarchical structure roughly consisting of tubes, sticks, beads, and plates organized at different hierarchical length scales (16). Despite the fact that we cannot directly probe sub-micron displacements in tissues with current MR hardware, we can infer some of these tissue’s topological features at finer length scales. By measuring the average propagator over a range of diffusion times, and using models of fractal diffusion appropriate for each tissue type, we can extrapolate into a high-resolution regime, inferring architectural features of tissue too small to be probed directly by conventional PFG MR (17). Although more data intensive than conventional PFG MRI, these approaches may one day provide sensitive, early biomarkers of changes in tissue microstructure associated with disease and development. There are several newly proposed parameters that characterize the fractal structure of complex, hierarchically organized media, which possess many of the features of quantitative imaging parameters.