DWI explores the random motion of water molecules in the body. Water molecules held in a container outside the body are in constant random brownian motion. This uninhibited motion of water molecules is free diffusion. By contrast, the movement of water molecules in biologic tissues is restricted because their motion is modified and limited by interactions with cell membranes and macromolecules.

In biologic tissue, the DWI signal is derived from the motion of water molecules in the extracellular space, the intracellular space, and the intravascular space [4] (Fig. 1A, 1B). Not surprisingly, given a unit time, water molecules in the intravascular space will have a greater diffusion distance because of blood flow than those in the extracellular and intracellular spaces. Clearly, the contribution of intravascular water diffusion to the measured DWI signal can vary among tissues. In tumors showing increased vascularity, the contribution of intravascular water diffusion to the MR signal may account for a significant proportion [5].

The degree of restriction to water diffusion in biologic tissue is inversely correlated to the tissue cellularity and the integrity of cell membranes [6–9]. The motion of water molecules is more restricted in tissues with a high cellular density associated with numerous intact cell membranes (e.g., tumor tissue). The lipophilic cell membranes act as barriers to motion of water molecules in both the extracellular and intracellular spaces. By contrast, in areas of low cellularity or where the cellular membrane has been breached, the motion of water molecules is less restricted. A less cellular environment provides a larger extracellular space for diffusion of water molecules, and these molecules may also freely transgress defective cell membranes to move from the extracellular into the intracellular compartment (Fig. 1A, 1B).

Stejskal and Tanner [10] described an MR experiment that could be applied to the detection and quantification of water diffusion in vivo. They adapted a standard T2-weighted spin-echo sequence by applying a symmetric pair of diffusion-sensitizing (bipolar) gradients around the 180° refocusing pulse. That approach is now the basis of many DWI sequences in clinical use today. Static molecules acquire phase information from the first diffusion gradient, but information will be rephased by the second diffusion gradient without a significant change in the measured signal intensity. By comparison, moving water molecules acquire different phase information from the first gradient, but because of their motion, their signal will not be completely rephased by the second gradient, thus leading to a signal loss (Fig. 2). Hence, the motion of water molecules is detected as attenuation of the measured signal intensity at DWI. The degree of water motion has been found to be proportional to the degree of signal attenuation.

The sensitivity of the DWI sequence to water motion can be varied by changing the gradient amplitude, the duration of the applied gradient, and the time interval between the paired gradients. On clinical MR scanners, the diffusion sensitivity is easily varied by changing the parameter known as the “b value,” which is proportional to these three factors. When the b value is changed, it is usually the gradient amplitude, rather than the duration or time intervals between gradients, that is altered.

Water molecules with a large degree of motion or a great diffusion distance (e.g., within the intravascular space) will show signal attenuation with small b values (e.g., b = 50–100 s/mm²). By contrast, large b values (e.g., b = 1,000 s/mm²) are usually required to perceive slow-moving water molecules or small diffusion distances because these show more gradual signal attenuation with increasing b values.

The root mean square displacement of water molecules that occurs during a diffusion measurement is estimated to be approximately 8 μm [11]. By comparison, the mean size of cells in the human body measures about 10 μm. Hence, the information provided by DWI reflects changes that are occurring at a cellular spatial scale. For this reason, DWI is perceived as an advantageous tool for evaluating changes in the tumor microenvironment, both before and after treatment.

DWI is typically performed using at least two b values (e.g., b = 0 s/mm² and other b values from 0 to 1,000 s/mm²) to enable meaningful interpretation. Generally, the larger the b value, the greater the degree of signal attenuation from water molecules. By observing the relative attenuation of signal intensity on images obtained at different b values, tissue characterization based on differences in water diffusion becomes possible. For example, in a heterogeneous tumor, the more cystic or necrotic fraction of the tumor will show greater signal attenuation on high b-value images because water diffusion is less restricted. By contrast, the more cellular solid tumor areas will continue to show relatively high signal intensity (Fig. 3A, 3B, 3C). Visual assessment of the relative tissue signal attenuation at DWI is being applied for tumor detection, tumor characterization, and the evaluation of treatment response in patients with cancer.

When diffusion measurements are being performed, the direction of water diffusion along the three orthogonal directions of the magnet (phase select, frequency select, and slice select) can be assessed independently by applying diffusion gradients in each of these directions. DW images that are the sum of the directionally acquired DW images are known as trace or index DW images.

One of the pitfalls of visual assessment of directional or index DW images is that the signal intensity observed depends on both water diffusion and the T2 relaxation time. Consequently, an area with a very long T2 relaxation time may remain high signal at DWI and be mistaken for restricted diffusion. This is known as the “T2 shine-through” effect (Fig. 4A, 4B, 4C). This effect can be sometimes reduced by the choice of an appropriate TE (a short one) and b value (a large one), but it cannot be easily avoided. The relative contribution of T2 signal intensity to DW images is a potential source of error in image interpretation and can limit their usefulness when comparing results among studies performed using different imaging protocols. A simple solution to the problem of T2 shine-through can be found using the exponential image. The exponential image is formed by taking the ratio of a DW image divided by an unweighted image (b = 0 s/mm²) from the same image series and slice position.

Another phenomenon that can be encountered is that of diffusion anisotropy. Diffusion anisotropy refers to unequal directional diffusion, which occurs as a result of tissue or structural organization. A good example of diffusion anisotropy is seen along the white matter tracts of the internal capsule in the brain. The diffusion motion appears relatively free in the head–foot direction along the long axis of the white matter tracts, but appears restricted in the anteroposterior and right–left directions across the neuronal fibers. In the assessment of tumors, diffusion anisotropy can help to identify tumor invasion of adjacent structures [12]. However, diffusion in tumors is usually isotropic because malignant cells typically grow in a disorganized fashion.

By performing DWI using different b values, quantitative analysis is possible. This analysis is usually performed at a push of a button on the scanner or workstation that results in the calculation of the apparent diffusion coefficient (ADC) (Appendix 1). One simplified method of visualizing this process is by considering the signal attenuation of a tissue with increasing b values. By plotting the logarithm of the relative signal intensity of the tissue on the y-axis against the b values on the x-axis, a line can be fitted through the plots (exponential function). The slope of the line describes the ADC (Fig. 5A, 5B, 5C). Not surprisingly, the fit can be improved by using more b values to reduce the error in ADC calculation.

The ADC is independent of magnetic field strength and can overcome the effects of T2 shine-through, thus allowing more meaningful comparison of results. The ADC is calculated for each pixel of the image and is displayed as a parametric map. By drawing regions of interests on these maps, the ADCs of different tissues can be derived.

Areas of restricted diffusion in highly cellular areas show low ADC values compared with less cellular areas that return higher ADC values. At this point, it is important to mention that although areas of restricted diffusion will appear to be higher in signal intensity on the directional or index DW images, these areas will appear as low-signal-intensity areas (opposite to DW images) on the ADC map (Fig. 5A, 5B, 5C).

However, it is simplistic to assume that the attenuation of the signal intensity with increasing b values occurs linearly in tissues. In the liver, the signal intensity can be seen initially to attenuate rapidly with low b values, followed by a more gradual signal reduction (Fig. 6). The initial reduction in signal intensity is believed to be due to vascular capillary perfusion, where the signal from fast-moving water molecules is rapidly attenuated by low diffusion weighting [4]. Thus, diffusion-weighted data acquired over a range of b values that includes low b values (< 50 s/mm²) are sensitive to signal attenuation from perfusion.

From the previous discussion, it stands to reason that deriving a flow-sensitive ADC value and a flow-insensitive ADC value is possible [4]. An exponential function fitted across all the b values describes the conventional ADC (flow sensitive). However, an exponential function fitted through only the high b values (e.g., b = 150–500 s/mm²) can be used to describe a flow-insensitive ADC. The flow-insensitive ADC may potentially provide a more accurate estimate of the cellularity of the tumor microenvironment by minimizing any vascular contributions (Fig. 6). As a further refinement, a biexponential function may be used to calculate conventional ADC from DWI that includes low b values [4].

In the brain, the presence of highly organized anatomic structures allows a particular quantitative diffusion technique, diffusion tensor imaging, to be used to advantage. Diffusion tensor imaging yields information about both the rate and the direction of water diffusion by sampling water motion in at least six directions. The potential application of diffusion tensor imaging in organs with discernible structural organization, such as the prostate and kidneys, is currently being evaluated [13–15].

Tumors are frequently more cellular than the tissue from which they originate and thus appear to be of relatively high signal intensity (restricted diffusion) at DWI.

DWI is being applied for the detection of liver metastases. In the liver, low b-value images (e.g., b = 50–150 s/mm²) that suppress the high-signal flow from the hepatic vessels, resulting in black blood images, have been found to be useful for lesion detection [24] (Fig. 8A, 8B, 8C, 8D, 8E). Metastases appear as high-signal-intensity foci at DWI.

Nasu et al. [25] found that DWI was more accurate than superparamagnetic iron oxide (SPIO)–enhanced MRI for the detection of liver metastases. In that study, the sensitivity and specificity of SPIO-enhanced MRI were 66% and 90%, respectively. By comparison, DWI was found to have a higher sensitivity of 82% and specificity of 94%. In another study, DWI was found to have a similarly high sensitivity of 86% and specificity of 94% for the detection of colorectal hepatic metastases [26].

Some of the challenges encountered in DWI of the liver are cardiac motion and susceptibility artifacts that can obscure or diminish visualization of the left lobe (Fig. 7A, 7B, 7C, 7D). The susceptibility effects may result from air in the adjacent stomach or colon. Artifacts resulting from cardiac motion can be reduced by triggered acquisition by ECG or a peripheral pulse unit, thus improving image quality and signal-to-noise ratio in the left lobe of the liver [27, 28]. Images may also be acquired with the aid of respiratory triggering to minimize inadvertent breathing motion. However, these techniques increase the image acquisition time, which can render the examination more susceptible to bulk motion.

Tumors differ in their cellularity, and this difference may reflect their histologic composition and biologic aggressiveness. The use of DWI for tumor characterization was first shown in brain tumors. The interested reader might like to refer to a more in-depth account of the use of DWI in the evaluation of intracranial tumors [29], which is beyond the scope of this article.

To characterize lesions in the liver using DWI, b values ranging between 0 and 500 s/mm² are appropriate [25, 30]. Qualitative visual assessment can help to distinguish cystic from solid lesions (Fig. 9A, 9B, 9C, 9D, 9E, 9F). However, it is often difficult to distinguish different types of solid lesions from one another in the liver by visual assessment alone. For example, a hemangioma will exhibit restricted diffusion and can mimic the appearance of a metastasis at DWI (Fig. 9A, 9B, 9C, 9D, 9E, 9F).

Using quantitative evaluation, investigators have found that benign liver lesions, such as cysts and hemangiomas, have higher mean ADC values (e.g., 2.45 × 10^–3 mm²/s) than malignant lesions, such as metastases and hepatocellular carcinoma (e.g., 1.08 × 10^–3 mm²/s) [31, 32]. The ADC has also been used to distinguish abscesses, which have low ADC values, from cystic and necrotic metastases, which have higher ADC values [33]. However, although the ADC values differ in summary statistics between benign and malignant lesions, using an individual ADC value to characterize lesions prospectively can be difficult. This difficulty is due to the considerable overlap in the ADC values of benign and malignant abnormalities. The wide range of ADCs in tumors may be explained by biologic variations of tumors, which become even more obvious across studies. For example, although some researchers have reported that metastases have lower ADC values than liver [31], others have found the ADCs of metastases to be higher than that of liver [34, 35]. Another possible reason for the wide range of ADC values may be errors in the measurements and image analysis. Hence, for the time being, DWI of the liver for lesion characterization should be combined with all available imaging to make the optimal assessment.

In the abdomen, DWI has also been applied to characterize focal renal lesions [18, 36]. Because of the organization of the renal tubules, water diffusion in the normal kidney is anisotropic [13, 37]. DWI in the kidney can readily distinguish between cystic from solid renal lesions [18]. However, it is not yet possible to confidently distinguish malignant from benign renal neoplasms on the basis of qualitative assessment or ADC measurements [18]. Elsewhere, the ADCs of malignant breast lesions have been found to be lower than the ADCs associated with benign diseases [6, 38, 39], and DWI was able to distinguish cystic soft-tissue sarcomas from solid types [40]. More recently, DWI has also been found to be useful in detecting colorectal carcinoma [41] and showing cystic lesions of the pancreas [42] and ovaries [43].

In prostate cancer, differentiating tumor from other causes of a low-signal-intensity lesion in the prostate gland is difficult on conventional T2-weighted MRI. Recently, DWI has shown potential for tumor identification [44, 45]. The normal central gland of the prostate has a lower ADC than the peripheral zone [44, 45]. Prostate cancers, which appear as low-signal-intensity foci on ADC maps, typically show lower ADC values than the peripheral zone and the transitional zone and central gland [44, 45]. However, there is significant overlap in the ADC values of prostate cancer and benign prostate changes [44]. Interestingly, investigators have recently observed that low ADC in the central gland and transitional zone accompanied by the loss of glandular anisotropy is more suggestive of tumor than glandular hyperplasia [46].

In the spine, the ability to distinguish a malignant from a nonmalignant cause of vertebral collapse remains challenging. Malignant vertebral infiltration and fracture frequently appear as high-signal-intensity areas on DWI compared with nonmalignant causes [22, 47–51]. A sensitivity of 42–100% and specificity of 92–94% have been reported using visual qualitative assessment [22, 47, 48], although one study found DWI to be unhelpful [52]. By quantitative analysis, malignant vertebral body infiltration and vertebral body fracture return lower ADC values than benign causes [48, 49, 53–56]. A clear separation of ADC measurements between malignant and benign groups has been observed [48, 53], but significant overlap has also been reported [56]. Unfortunately, the threshold ADC values that can be applied to discriminate between benign and malignant vertebral disease vary among studies because of differences in the imaging sequences and parameters. Hence, the reader should exercise caution when adopting these criteria for clinical practice. Furthermore, inflammatory conditions, such as tuberculous infection [49] and osteomyelitis [57], can mimic malignant disease on DWI.

One potential use of DWI is in the identification of recurrent or residual tumor after treatment. A recent pilot study in patients with laryngeal cancer showed the potential of the technique to differentiate residual or recurrent tumor from postchemoradiation changes [58]. What was particularly interesting was the fact that DWI could be used to correctly identify a patient as showing posttreatment changes even when PET suggested recurrent disease [58]. Clearly, the promising utility of DWI for this purpose requires further evaluation.

There is increasing interest in the application of DWI for detecting tumor response. Effective anticancer treatment results in tumor lysis, loss of cell membrane integrity, increased extracellular space, and, therefore, an increase in water diffusion [59, 60]. The proposed mechanisms in the variation of tumor ADC after the initiation of treatment are summarized in Figure 10 [59].

However, this scheme does not consider the contribution of intravascular perfusion to the diffusion measurement, which may be substantial in a tumor [61]. Hence, therapies that are targeted against tumor vasculature may also result in a reduction in the ADC, especially when the DW images are acquired using low b values, which are sensitive to vascular perfusion effects [62].

The results of animal studies have confirmed that after the initiation of chemotherapy, radiation therapy, or novel therapy, an increase in the ADC value may be observed in those responding to treatment [62]. Furthermore, treatment effects can be observed within the first 24 hours after initiating treatment [5, 62] due to cell swelling, which results in a transient decrease in the ADC [5, 62].

In human studies of vertebral metastases [50] and brain tumors [63], investigators have reported that treatment response can be observed as a reduction in high signal intensity at DWI. However, such visual appraisal may be confounded by T2 shine-through effects. Using ADC measurements, researchers who have studied hepatocellular carcinoma [64], cerebral gliomas [65], and soft-tissue sarcoma [66] have found that individuals who respond to treatment show a significant rise in the ADC values after therapy. In a recent study of colorectal hepatic metastases, an increase in ADC was observed in patients with at least a partial response to treatment [67]; however, an ADC increase was not observed in the nonresponders [67].

One of the most intriguing findings associated with the use of DWI in cancer patients has been that ADC measurements appear to be able to predict the response of tumor to chemotherapy and radiation treatment. Table 2 summarizes the findings of selected clinical DWI studies with results that appear to document this phenomenon.

Studies in rectal carcinoma [68, 69], cerebral gliomas [59, 63, 70], and colorectal hepatic metastases [67] have shown that cellular tumor with low baseline pretreatment ADC values respond better to chemotherapy or radiation treatment than tumors that exhibit high pretreatment ADC values. One possible explanation is that tumors with high pretreatment ADC values are likely to be more necrotic than those with low values. Necrotic tumors frequently are hypoxic, acidotic, and poorly perfused, leading to diminished sensitivity to chemotherapy and to radiation therapy.

Studies in cerebral gliomas [59, 63, 71] and breast carcinoma [72] have also shown that an early increase in the ADC after commencing treatment was predictive of better treatment outcome. In two of the studies [71, 72], researchers found that an increase in the ADC within 1 week of initiating treatment was predictive of at least a partial response, with response being defined by tumor size reduction at the end of therapy. The increase in the ADC preceded any reduction in tumor size. The use of ADC to evaluate and predict response has also been assessed in a number of animal studies [62, 73–79] (Table 3).

Many of the clinical studies evaluating DWI for assessing treatment response have been performed in relatively small numbers of patients. Nevertheless, the body of evidence suggests that ADC measurement is a potentially useful tool that provides unique prognostic information and should be more widely investigated in large clinical studies in the future.

Whole-body DWI is a recently developed application of DWI that, as previously described, is performed using a STIR EPI diffusion-weighted technique with a high b value of 1,000 s/mm² for background suppression. By performing imaging at multiple stations in the body, a composite image of the whole body can be constructed. The images are processed using maximum intensity projection and are usually displayed using a reversed black-and-white gray scale. Signals from normal tissue such as blood vessels, fat, muscle, and bowel are suppressed. However, other normal structures such as the spleen, prostate, testes, ovaries, endometrium, and spinal cord remain visible [23]. Areas showing restricted diffusion—for example, highly cellular lymph nodes—are strikingly depicted (Fig. 11A, 11B). Not surprisingly, the technique has been applied to evaluate lymphadenopathy in patients with lymphoma and other cancers. Exquisite images displaying small foci of tumors within the abdomen or peritoneum have also been presented in the literature using this technique [23].

The true utility of whole-body imaging requires further validation. One potential application of DWI is the detection of tumors. This application would be analogous to using whole-body STIR technique for tumor detection. However, because DWI can detect abnormalities on the basis of tissue cellularity, it is unclear whether abnormalities depicted using this technique are necessarily malignant. There is currently a great deal of interest in the development of whole-body DWI with a background-suppression technique. However, more research is needed to establish the sensitivity and specificity of the technique to detect nodal malignancy. Another challenge that must be overcome for DWI to be used for tumor detection is accurate quantification of the ADC.