AIM: To evaluate the role of diffusion-weighted imaging (DWI) in the detection of focal liver lesions (FLLs), using a conventional magnetic resonance imaging (MRI) protocol.

METHODS: Fifty-two patients (22 males, average age 55.6 years, range: 25-82 years), studied using a 1.5 Tesla magnetic resonance scanner, were retrospectively analyzed; detection of FLLs was evaluated by considering the number of lesions observed with the following sequences: (1) respiratory-triggered diffusion-weighted single-shot echo-planar (DW SS-EP) sequences; (2) fat-suppressed fast spin-echo (fs-FSE) T2 weighted sequences; (3) steady-state free precession (SSFP) images; and (4) dynamic triphasic gadolinium-enhanced images, acquired with three-dimensional fast spoiled gradient-echo (3D FSPGR). Two radiologists independently reviewed the images: they were blinded to their respective reports. DW SS-EP sequences were compared to fs-FSE, SSFP and dynamic gadolinium-enhanced acquisitions using a t-test. Pairs were compared for the detection of: (1) all FLLs; (2) benign FLLs; (3) malignant FLLs; (4) metastases; and (5) hepatocellular carcinoma (HCC).

RESULTS: Interobserver agreement was very good (weighted κ = 0.926, CI = 0.880-0.971); on the consensus reading, 277 FLLs were detected. In the comparison with fs-FSE, DW SS-EP sequences had a significantly higher score in the detection of all FLLs, benign FLLs, malignant FLLs and metastases; no statistical difference was observed in the detection of hepatocellular carcinoma (HCCs). In the comparison with SSFP sequences, DW SS-EP had significantly higher scores (P < 0.05) in the detection of all lesions, benign lesions, malignant lesions, metastases and HCC. All FLLs were better detected by dynamic 3D FSGR enhanced acquisition, with P = 0.0023 for reader 1 and P = 0.0086 for reader 2 in the comparison with DW SS-EP sequences; with reference to benign FLLs, DW SS-EP showed lower values than 3D FSPGR enhanced acquisition (P < 0.05). No statistical differences were observed in the detection of malignant lesions and metastases; considering HCCs, a very slight difference was reported by reader 1 (P = 0.049), whereas no difference was found by reader 2 (P = 0.06).

CONCLUSION: In lesion detection, DWI had higher scores than T2 sequences; considering malignant FLLs, no statistical difference was observed between DWI and dynamic gadolinium images.

In the past twenty years, several magnetic resonance imaging (MRI) diagnostic techniques have gradually been introduced in the study of focal liver lesions (FLLs), increasing to the greatest possible extent the high contrast resolution inherent in the method. For instance, long echo-time T2 weighted sequences are widely used to characterize fluid-filled cystic lesions or hemangiomas; fat-containing lesions (adenomas, atypical focal nodular hyperplasia) can be diagnosed in cases of signal decay in out-of-phase T1-weighted spoiled gradient echo sequences[1]. More recently a gradual introduction of new hepato-specific agents - gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid (Gd-EOB-DTPA, Primovist, Bayer Schering) and gadopentate dimeglumine (Gd-BOPTA, Multihance, Bracco Imaging) has significantly increased the diagnostic potential of MRI, especially in the preoperative management of patients with liver metastases[2,3]. In addition, three-dimensional gradient echo imaging allows a dynamic study of the liver, evaluating the behavior of FLLs in the arterial and portal phase after contrast administration[4]. In view of these considerations, non-invasive characterization of FLLs is a current diagnostic challenge.

The introduction of diffusion weighted imaging (DWI) in liver MRI fuelled high expectations with several encouraging studies[5-7]. High values of b - required for characterization - cannot be used to identify FLLs. It has been shown that the use of diffusion weighted single-shot echo-planar (DW SS-EP) sequences with a low b-value is important for the detection of hepatic lesions, especially for small-sized lesions. The acquisition of “black blood” images with DW SS-EP sequences makes it possible to easily differentiate vessels from focal small-sized lesions in the liver[8,9]. Parallel techniques reduce artifacts, such as “blurring” or magnetic susceptibility artifacts, increasing the signal-to-noise ratio (SNR)[10,11]; moreover, respiratory-triggered sequences offer a better possibility of SNR compared to breath-hold techniques[12]. Many studies have recently started to compare DWI to T1-weighted acquisition in the detection and characterization of liver lesions[13-15].

In diffusion imaging of the liver, the first difficulty concerns the choice of the appropriate b-value, which means the degree of weighting in diffusion. High b-values cannot be used because of the low SNR they determine: for these reasons, in this study we retrospectively evaluated the ability to detect lesions using relatively low b-value DWI (b-values extended from b = 0-10 to b = 0-500) in a routine liver MRI protocol. The characterization of FLLs was not considered in our work, because the role of DWI in this regard has already been widely debated in previous studies[5,15], and seems to be limited by overlapping[16].

In the detection of FLLs, diffusion weighted sequences were compared to fat suppressed fast spin-echo (fs-FSE) T2-weighted sequences, steady state free-precession (SSFP) sequences and dynamic triphasic gadolinium-enhanced acquisition, acquired with 3D fast spoiled gradient echo (FSPGR), emphasizing their advantages and limits.

Interobserver agreement was very good (weighted κ = 0.923, CI = 0.873-0.972). Detection rates for all FLLs, benign lesions, malignant lesions, metastases and HCC are shown in Table 2.

DW SS-EP sequences vs fs-FSE

DW SS-EP sequences showed a higher score when compared with conventional breath-hold T2-weighted sequence fs-FSE, with P = 0.001 for reader 1 and P = 0.0019 for reader 2 (Figure 1A). Considering benign FLLs (Figure 1B), DW SS-EP sequences showed higher values than fs-FSE sequences (P = 0.00 017 for reader 1 and P = 0.01 for reader 2) and SSFP sequences (P = 0.00 013 for reader 1 and P = 0.02 for reader 2); in the evaluation of malignant lesions (Figure 1C), DW SS-EP showed higher scores than fs-FSE only for reader 1 (P = 0.023), whereas no statistical difference was observed by reader 2 (P = 0.053). In the detection of metastases (Figure 1D), a significant statistical difference was observed for both readers, with P = 0.026 for both readers; no statistical difference was found in the detection of HCC (Figure 1E).

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Histograms comparing the different sequences (A-E). Histograms showing the comparison between the mean values of lesions detected among different sequences, for the classes of focal liver lesions (FLLs) [all lesions, benign lesions, malignant lesions, metastases, hepatocellular carcinomas (HCCs)]. I= reader 1; II = reader 2; Tri-Gd T1: Triphasic gadolinium T1-weighted sequence; fs-FSE: Fat suppressed fast spin-echo sequence; SSFP: Steady state free precession; DWI: Diffusion weighted imaging.

In recent years, some studies have emphasized the role of DWI in the characterization and detection of FLLs[5-7,9-12,19-21]. Furthermore, DWI could provide an additional tool in oncological patients, where a correct assessment of the number and localization of FLLs is essential in choosing suitable treatment. In fact, DWI provides better detection of FLLs than other T2-conventional sequences, and our results confirm what has recently been published in the literature. For instance, in a previous study by Bruegel et al[10], the sensitivity of DWI was higher than standard breath-hold and respiratory-triggered T2 weighted sequences in the diagnosis of 66 small hepatic metastatic lesions (sensitivity value of 0.85 and 0.26-0.44, respectively)[10]. In a study by Holzapfel et al[21], the detection of small FLLs seemed to be significantly increased by respiratory-triggered diffusion sequences, which produced better image quality because of high spatial resolution and an adequate SNR[22]. In the detection of metastases, when DWI was compared to fs-FSE sequences and to SSFP, we found a significantly higher number of lesions identified by both readers using DWI; the highest detection rate - shown by DWI in the detection of FLLs - was probably due to the ‘black blood effect’ of diffusion-weighted images: this property makes it easier to distinguish small FLLs from hepatic vessels, which may determine false-positive pseudolesions on T2-weighted turbo spin-echo images.

In view of its capability to detect FLLs, DWI has been widely introduced in liver imaging, and many studies have recently emphasized its additional values in liver imaging in order to detect malignant FLLs[13-15]. In this regard, a routine MRI protocol should ideally use all possible strategies, including many un-enhanced scans: in and out-of-phase T1-weighted sequences, T2-weighted FSE sequences - even with spatial fat saturation or different echo-time, white-blood sequences for morphologic assessment of liver anatomy such as steady-state sequences, and SSFSE sequences to emphasize T2-content; dynamic multiphase gadolinium-enhanced acquisitions are necessary to subsequently evaluate FLLs and to study their behavior in the arterial, portal and late phases. For this reason, the role of DWI in a liver MRI protocol should be assessed comparing the diffusion sequences not only to the conventional T2 sequences, but it must also include the dynamic enhanced gadolinium acquisitions.

In a study by Löwenthal et al[15] - considering only the detection of benign FLLs - a statistical difference in the comparison between DWI and hepatobiliary phase (MR-late) Gd-EOB-DTPA-enhanced images was reported; no difference was observed between DWI and dynamic (MR-dyn) Gd-EOB-DTPA-enhanced images, unlike the data reported in our comparison. This may be explained by the fact that DWI is less reliable than dynamic enhanced gadolinium FSPGR acquisitions, being much more sensitive to magnetic susceptibility artifacts. For instance, in our study, 6 cases of FNHs were missed by both readers evaluating only the diffusion-weighted images, probably due to their location in the parenchyma of the left lobe, near the gastric area: in this area DWI (Figure 2) could be affected by magnetic susceptibility or motion artifacts related to cardiac and gastric wall.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Focal nodular hyperplasia found in the left liver lobe (arrow). A: A round-shaped solid lesion (arrow) is depicted in the left liver lobe (in the II segment); B, C: The lesion (arrow) appears homogeneously hyperintense in the arterial phase, and remains slightly hyperintense in the portal and in the equilibrium phase; D: The lesion was missed on the diffusion image by both readers, probably due to its location near the gastric wall, along the liver surface; E: In the hepato-specific phase, the lesion (arrow) shows uptake of gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid; a diagnosis of focal nodular hyperplasia was suggested due to the dynamic behavior observed after contrast administration.

On analyzing the comparison using the t-test, no statistical difference between the number of lesions identified by DWI and dynamic enhanced gadolinium acquisition was observed, and DWI could be considered a valid diagnostic tool in the detection of malignant lesions in a liver study, even if they are characterized by a lower sensitivity in comparison to the dynamic acquisition. The high detection rate using DWI is especially observed in the evaluation of metastases, as confirmed in other studies[14]. The good results obtained using DWI in our study may be explained by the application of respiratory triggering, which produced better image quality because of high spatial resolution and an adequate SNR[12]; use of the respiratory-triggered modality in the diffusion images increases the spatial resolution, but requires a longer time for the acquisition. In addition, the use of low b-value diffusion sequences, acquired by parallel techniques, reduces artifacts such as “blurring” or magnetic susceptibility artifacts, increasing the SNR[10,11].

In a recent study by Kenis et al[13], DWI was used in the staging of oncological disease in patients with impaired renal function; in this study, MR showed a significant additional capability in the detection of metastases when considering DWI and gadolinium MRI together[13]. Kenis et al[13] analyzed three image sets: unenhanced T1 and T2 acquisitions/gadolinium-enhanced T1, DWI and a combination set; all observers reported high sensitivity values using the combination set of images; as a consequence, diagnostic accuracy increased significantly when DWI was added to Gd-MRI.

DWI is limited in the detection of HCC, with a mean detection rate (62.5%) lower than 3D FSPGR enhanced images (100%). Five HCC lesions missed by both readers were located in a subdiaphragmatic liver segment close to the heart or gastric wall (Figure 3): again, the lower diagnostic accuracy of DWI was probably related to the presence of magnetic susceptibility or motion artifacts in these locations, and even if the introduction of triggered-acquisition increased their image quality, we suggest that the evaluation of these areas could be limited. Similarly, magnetic susceptibility artifacts caused by colonic loop - often located very close to the caudal portion of the right liver - reduces imaging quality on diffusion images, with low signal intensity of parenchyma and loss of liver profile: one metastasis missed by both readers was located in the caudal portion of the IV segment, near to the colonic wall (Figure 4). In addition, lower scores obtained with DWI in our study - regarding the detection of HCCs - may also be explained by the different signal intensity observed in these lesions: in fact, in a recent study by Kim et al[14] they were isointense or hyperintense to the liver[14]. In a cirrhotic liver, HCCs may show the same signal intensity as the surrounding parenchyma, involved in a chronic fibrotic process, and as a consequence the detection and characterization of HCCs may be difficult[14].

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Hepatocellular carcinoma in the left liver lobe. A: Hypervascular lesions depicted (arrows) in the arterial phase; B: Lesions show wash-out in the venous phase (arrows), suggesting the diagnosis of hepatocellular carcinomas; C: On diffusion imaging, acquired at b10 value, lesions were not detected by the readers, due to their location near the gastric wall.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Metastasis in the IV segment. A: A metastatic lesion in colon carcinoma is depicted in the most caudal area of the IV liver segment, it was well depicted in the fat-suppressed fast spin-echo T2-weighted sequence (arrow); B: In the gadolinium-enhanced acquisition (arrow); C: The lesion is poorly represented in the diffusion-weighted imaging, covered by magnetic susceptibility artifacts due to the adjacent intestinal loop (arrows); D: Delayed imaging after gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid confirmed the metastatic lesion (arrow).

Although some artifacts may affect the imaging quality of diffusion images, the highest detection rate observed in a liver MRI protocol suggest that they can easily help radiologists in the detection of FLLs, and in particular, in the evaluation of metastases. These lesions are often reported with a larger size in the diffusion images, in comparison to the measurements observed by enhanced acquisitions. This larger size is better depicted in the follow-up of oncological patients, where lesions seem to be more reduced than those reported by gradient T1-weighted acquisition.

The main limitations of our work include the small number of histological proofs for FLLs: this may be explained by the fact that we retrospectively analyzed our study population and had many difficulties in collecting the patients’ data. In addition, we discarded many exams when the image quality was damaged by artifacts, or in cases of incomplete liver MRI study.

There was no comparison between DW SS-EP and FSPGR images obtained in the hepato-specific phase; this may represent a limitation, considering the high potential value of liver-specific contrast assessed in the literature, especially in the evaluation of HCC[22]. Unfortunately, the number of exams including the hepato-specific phase in our study was not consistent and adequate for statistical analysis.

In conclusion, DWI may be very helpful to radiologists in a liver MRI protocol, increasing the detection of FLLs; considering the data reported in our study, the reliability of DWI is slightly lower than dynamic 3D FSPGR gadolinium-enhanced acquisition, but higher than conventional T2-weighted sequences. Regarding malignant FLLs, no statistical difference was found between DWI and conventional dynamic gadolinium images in our study; nevertheless, we feel contrast administration is mandatory to ensure the absence of malignant liver diseases, especially HCCs.