The possibility of detecting and measuring the diffusivity of water with Nuclear Magnetic Resonance (NMR) methods was reported in the early work of pioneers of NMR in the early fifties (e.g. E. L. Hahn, Phys. Rev. 80, 580–594 (1950). In the seventies the first spectroscopic investigations of diffusion in biological tissues took place, followed in the mid eighties by the first clinical demonstration of diffusion weighted MRI (DWI). The investigation of diffusion in anisotropic and heterogeneous tissues has been facilitated by the introduction of diffusion tensor MRI (DTI). From diffusion tensor data one can compute quantities that characterize specific features of the diffusion process, such as the principal diffusivities (eigenvalues of D), the trace of the diffusion tensor (Trace(D)), indices of diffusion anisotropy, and the principal directions of diffusion (eigenvectors of D). Diffusion tensor MRI (DTI) has been extensively used for probing features related to composition, microstructure, and organization of tissues in the brain and other organs of human subjects. There is a large amount of data indicating that DT-MRI could improve the clinical assessment of several neurological and psychiatric disorders. Promising clinical applications of DT-MRI have been proposed also in organs other than the brain, such as kidney, liver, prostate, skeletal muscle. Diffusion MRI methods can be used to infer the trajectories of white matter fibers in the brain. Since the initial DTI “tractography” studies published more than 10 years ago, several more sophisticated methods have been proposed generating a lot of enthusiasm in the neuroscience community in the hope that these tools could help elucidating anatomical connectivity in the central nervous system. Finally, in recent years, we have seen a renewed interest in probing diffusion with very strong diffusion sensitization and very large angular resolution for sampling, allowing the application of more sophisticated analysis models aimed at characterizing features of the tissue that are not properly investigated with the tensor model.

However, despite the large body of clinical and experimental studies published, quantitative Diffusion MRI has still very little penetration into clinical practice. In this talk we will review some of the obstacles that have hindered a larger dissemination of Diffusion MRI and the most promising solutions that have been proposed to address them. In particular we identify the following aspects that affect the clinical applicability of diffusion MRI:

1) Quality of Diffusion MRI acquisitions. The quality of Diffusion MRI is generally poor compared to that of other structural MRI acquisitions because clinical DWIs are acquired using single shot echo planar imaging (SS-EPI). SS-EPI has the advantage of being efficient (high SNR per unit time) and more immune from motion related-ghosting than segmented acquisition. However, the spatial resolution and anatomical accuracy of EPI is suboptimal in most clinical scanners. We will discuss the impact of EPI artifacts on DT-MRI and review strategies for correcting residual EPI related distortions via non-linear image registration. A number of excellent strategies for controlling and correcting EPI distortions, eddy current related distortions, and image misregistration caused by subject motion have been proposed in the last few years.

2) Artifacts affecting accuracy and reproducibility of Diffusion MRI. Although DT-MRI is a quantitative technique (i.e. it measures a physical quantity that is reported in absolute units), several factors adversely affect the accuracy and precision of DTI measures. Such factors can be broadly classified as originating from thermal noise, system induced artifacts, and physiological noise. Physiological noise originates from subject motion, cardiac pulsation, partial volume contamination from cerebral-spinal fluid, and, possibly, respiratory motion and blood flow induced pseudo-diffusion effects. These factors affect the reproducibility of clinical DTI scans and negatively impact clinical studies in several ways. The topological distribution of the effects of artifacts is not uniform troughout the brain, and in many cases is not known at the time of designing the study. Not knowing the overall variability of DTI measurements precludes computing the number of subjects necessary to be able to detect a given effect. Moreover, longitudinal data and data from different centers cannot be compared reliably. Sources of variability also act as confounds in assessing differences between different groups of subjects. For example, DT-MRI differences found between healthy controls and patients may be due to artifacts originating from physiological noise, such as heart rate and subject motion, rather than to true anatomical differences. There is clearly a need for improving the resolution, reliability, and overall quality of diffusion tensor MRI acquisitions. The challenge is to achieve this goal by maintaining a reasonably short scan time.

3) Biological specificity and Validation of Diffusion MRI. In general, understanding the relationship between a measured water diffusion pattern and the underlying histological features of the tissue is not simple. We lack a robust and comprehensive model that relates water diffusivity to specific biological features. Essentially, we have not proven convincingly that diffusion metrics can be used as specific biomarkers. Definitely additional studies in animal models will be helpful in claryfing the biomarker specificity of Diffusion MTI metrics, however, there may be situations the can not be disambiguated even after extensive histological investigation. For example increase diffusivity is observed in edematous but otherwise healthy tissue as well as in necrotic tissues when lysis of cellular elements occurs. Two very different conditions with a very similar Diffusion MRI signature. As a different example, the main problem in inferring white matter trajectories from diffusion MRI measurements is that the diffusion properties measured in a voxel are affected by the presence of a large quantity of axons. The measured diffusion displacement profile is essentially a voxel-averaged quantity which provides a good estimate of fiber orientation only if the axons are oriented collinearly. In heterogeneous tissue, inferring the intravoxel architecture of white matter becomes a complicated inverse problem which we believe is essentially unsolvable with the limited information gathered by Diffusion MRI.

Ultimately the penetration of a MRI technique into clinical practice is related to its ability to answer reliably, quickly, and inexpensively questions such as: Does this technique have good sensitivity and specificity in detecting disease; Is it useful for staging the disease? Does it provide metrics that can be considered biomarkers? Is it sensitive to hidden pathology not revealed by other techniques? Can it help differentiating between different clinical subgroups? Is there a relationship between changes in imaging parameters and clinical disability? Is this technique helpful in assessing an individual patient ? Would the information I gain with the examination alter treatment choices and/or provide prognostic information ? The current lack of widespread clinical application of Diffusion MRI does not imply that it this technique is inherently unable of to offer valuable information in a clinical setting, but it clearly indicates that we need to put more effort in overcoming standing obstacles.