Tomoelastography Distinguishes Noninvasively between Benign and Malignant Liver Lesions

Liver mechanical properties affect mass formation and treatment outcome (1–3). Specifically, each 1-kilopascal increase in liver stiffness increases the probability of hepatocellular carcinoma (HCC) by 4% (2) and the risk of tumor recurrence after hepatic resection by 16.3% (3). This is the rationale for using elastography to measure mechanical tissue properties of the liver as predictive markers of tumor progression (4).

Complex shear modulus G* measured by magnetic resonance elastography (MRE) provides a basic description of mechanical tissue properties. Stiffness is usually characterized by magnitude modulus (|G*|), while elasticity and viscosity are quantified by storage (real part of G*) and loss modulus (imaginary part of G*), respectively. Phase angle ϕ of the complex modulus indicates fluidity, equaling zero for pure solids and π/2 radians (rad) for pure fluids. Thus, fluidity is conceptually signifying the conversion of a solid into a fluid over a continuous range from 0 to π/2. Materials with ϕ < π/4 have predominantly solid and materials with ϕ > π/4 have predominantly fluid properties (5).

Fluidity is currently under investigation as a mechanical property defining the metastatic potential and aggressiveness of tumors (6). Despite many reports on the stiffness of liver tumors (4), little is known about how disease alters the solid–fluid properties of liver tissue. Multifrequency MRE shows that a power-law coefficient related to ϕ does not change with progression of hepatic fibrosis, while tissue stiffness increases (7). Interestingly, Garteiser and colleagues reported a changed ratio of loss-to-storage properties for liver tumors, suggesting increased fluidity due to carcinogenesis (8). MRE without considering solid–fluid tissue properties is moderately sensitive to malignancy in the liver (8, 9). Given (i) the altered loss modulus in liver tumors (8), (ii) the insensitivity of fluidity to liver fibrosis (7), and (iii) a moderate diagnostic performance of stiffness in identifying liver tumors (9), we hypothesize that fluidity may be a promising imaging biomarker unaffected by the stiffness of nontumorous liver tissue, which is normally altered by fibrosis. However, fluidity has never been explored in liver tumors; specifically, novel high-resolution multifrequency MRE with tomoelastographic data processing (10, 11) has never been used for this.

Here we use tomoelastography to investigate whether solid–fluid properties can differentiate hepatic tumors from nontumorous liver tissue and malignant from benign lesions. This study was performed to quantify the solid–fluid properties of both tumor and nontumorous liver by tomoelastography.

The study protocol conforms with the Declaration of Helsinki and was approved by the institutional ethics review board (EA1/261/12 and EA1/076/17). All patients gave written informed consent.

From March 2015 to August 2017, the electronic scheduling systems of the departments of liver surgery and radiology were screened for patients with focal liver lesions detected by previous imaging and without contraindications to MRI. Patients with prior treatment of the respective lesion were excluded. The final study cohort included 70 patients with a total of 105 malignant and 36 benign lesions, and 7 healthy volunteers. Diagnoses were 22 HCC, 12 cholangiocarcinoma, 71 metastasis, 11 hemangioma, 10 focal nodular hyperplasia (FNH), and 15 hepatic adenoma. Details are provided in Table 1 and Supplementary Fig. S1.

MRI was performed on a 1.5-Tesla Clinical Scanner (Magnetom Aera, Siemens) with a 12-channel phased-array surface coil and the spine-array coils integrated into the table. The MRE sequence and set-up are described in detail elsewhere (11). In brief, mechanical vibrations of 30, 40, 50, and 60 Hz harmonic frequency were induced with two posterior (0.6 bar) and two anterior-pneumatic actuators (0.4 bar) near the liver region (Supplementary Fig. S2). The three-dimensional wave field was acquired by using a single-shot, spin-echo echo-planar imaging sequence with flow-compensated motion-encoding gradients (MEG). Eight phase offsets over a full vibration period were recorded for all three Cartesian-motion field directions. Fifteen consecutive transverse slices with a field-of-view of 384 × 312 mm² (matrix size 128 × 104) and 3 × 3 × 5 mm³ resolution were acquired during free breathing. Further imaging parameters: echo time = 59 ms; repetition time = 2,050 ms; parallel imaging with GRAPPA-factor 2; MEG frequency of 43.48 Hz for 30, 40, and 50 Hz vibration frequencies and 44.88 Hz for 60 Hz vibration frequency; and MEG amplitude of 30 mT/m. Total MRE measurement time was approximately 3.5 minutes. More details about the tomoelastography setup are provided in Supplementary Fig. S2.

MRE data were processed with the tomoelastography pipeline based on multifrequency wavenumber recovery providing maps of shear-wave speed c (in m/s; ref. 10). Because c is deduced from the real part of wave numbers, it can be considered a surrogate of stiffness. Furthermore, we recovered the phase angle of the complex shear modulus, ϕ (range: 0–π/2). ϕ > π/4 indicates that loss modulus dominates over storage modulus (12). Both parameters, c and ϕ, were retrieved by multifrequency dual elasto-visco inversion programs publicly available at https://bioqic-apps.charite.de/downloads. Volumes of interest (VOI), drawn manually, were based on all MRE magnitude images covering tumor or liver and using the information of the lesion's anatomic location obtained by routine MRI. For nontumorous liver tissue VOIs were drawn excluding visible blood vessels.

Means were calculated for tumor and liver tissue. Group values were reported as mean ± SD. Different tumor entities in two subjects were considered as independent lesions. Statistical differences between tumor entities were analyzed by the Kruskal–Wallis test. A post hoc test with Bonferroni correction for multiple testing was performed. For separation of lesions and nontumorous liver tissue, ROCs were calculated from probabilities predicted by a general linear mixed-effect model (GLMM; ref. 13). The outcome variable was labeled as “liver” or “tumor.” We accounted for fixed effects (β) for either c or ϕ and for nonindependence of multiple tumors within one subject. ROC measures were compared by DeLong method (14).

ROC analysis was performed for characterization (distinguishing malignant from benign lesions). Youden index was calculated to derive optimal cut-off values of ϕ or c. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated. ROC analysis was repeated without hemangioma because this is easily detected by its strong T2-hyperintensity in MRI (15). Correlation of mechanical properties of tumor and liver was analyzed by Spearman rank correlation coefficient. Statistical analysis was performed in R (version 3.4.3; R-Foundation) using “lme4,” “pROC,” and “ggplot2” packages. P < 0.05 was considered statistically significant. All confidence intervals (CI) stated are for 95%.

Tomoelastography was successful in all participants, that is, there was no drop out due to technical failure. Participant characteristics and group-mean values of c and ϕ of tumors and liver tissue are summarized in Table 1. MRE maps of four representative cases of malignant lesions including two metastasis, one HCC, and one cholangiocarcinoma are shown in Fig. 1A–D. Figure 2A–D shows four representative cases of benign lesions including two hepatic adenoma, one FNH, and one hemangioma. More cases and additional imaging data are presented in Supplementary Figs. S3A–S3D and S4A and S4B.

Overall, c and ϕ were significantly larger in malignant lesions (including metastasis, HCC, and cholangiocarcinoma) than in nontumorous liver tissue (respectively, β = 3.62, 95% CI = 2.41–5.73, P < 0.001 and β = 10.64, 95% CI = 7.02–16.35, P < 0.001). GLMM-based ROC analysis showed a significantly larger area under the ROC curve (AUC) of 0.95 (95% CI = 0.92–0.98) for ϕ, compared with 0.88 (95% CI = 0.83–0.94) for c (P < 0.01). Separation of malignant lesions and liver by c was feasible, with a sensitivity of 91% (95% CI = 86–96) and specificity of 70% (95% CI = 59–81), while separation of malignant lesions and liver by ϕ revealed 92% sensitivity (95% CI = 87–97) and 85% specificity (95% CI = 76–93). Detailed GLMM results for tumor and liver separation based on c and ϕ are shown in Supplementary Table S1.

Benign lesions (including hemangioma, FNH, and hepatic adenoma) separated from nontumorous liver based on c (P = 0.03), while ϕ was not significantly different (P = 0.051). Corresponding AUC values were 0.71 (95% CI = 0.57–0.84) and 0.66 (95% CI = 0.51–0.80) for c and ϕ, respectively. However, ROC curves of c and ϕ did not differ statistically (P = 0.57).

Results of GLMM-based ROC analysis for separation are shown in Fig. 3A and Supplementary Table S2.

Tumor entities varied significantly in their c and ϕ properties (both P < 0.001). Post hoc tests revealed smaller c values for hepatic adenoma than for all malignant tumor entities (P < 0.001), but not compared with other benign entities (Fig. 3B). In addition, hepatic adenoma showed smaller ϕ values than malignant tumors did (P < 0.01 vs. HCC and cholangiocarcinoma, P < 0.001 vs. metastasis). Two patients with large FNH showed high c and ϕ values. Additional information can be found in Supplementary Fig. S4A and S4B. Statistical plots of group mean c and ϕ are shown in Fig. 3B.

A lesion-c threshold of 1.75 m/s allowed distinction between benign and malignant lesions with a very good sensitivity of 94% (95% CI = 87–100) and moderate specificity of 78% (95% CI = 56–94; AUC = 0.85, 95% CI = 0.72–0.98). Similarly, lesion-ϕ of 0.91 rad, which indicates fluid tissue properties, distinguished benign from malignant lesions with a good sensitivity of 83% (95% CI = 72–93) and moderate specificity of 78% (95% CI = 56–94; AUC = 0.86, 95% CI = 0.77–0.96).

In a total of 13 patients, excluding patients with hemangioma of vascular origin (n = 5) from our analysis, diagnostic accuracy increased when the same lesion thresholds of c = 1.75 m/s and ϕ = 0.91 rad were used. Without hemangioma, sensitivity and specificity of c were very good, at 94% (95% CI = 89–100), and good, at 85% (95% CI = 62–100; AUC = 0.88, 95% CI = 0.73–1.00), respectively, while those of ϕ were good, at 83% (95% CI = 72–93) and very good, at 92% (95% CI = 77–100; AUC = 0.88, 95% CI = 0.73–1.00). These results show that diagnostic performance increases when benign tumors of vascular origin are excluded from the characterization of hepatic lesions based on their mechanical properties.

We also characterized liver lesions on the basis of nontumorous liver tissue. Liver properties in patients with benign lesions were not different from those of healthy volunteers (all P > 0.05). Optimum cutoffs for discrimination of embedded benign tumors from embedded malignant tumors were c = 1.5 m/s and ϕ = 0.60 rad, the latter being in the range indicating solid tissue properties. Sensitivity and specificity for liver-c were, respectively, moderate, 72% (95% CI = 59–83), and good, at 88% (95% CI = 71–100; AUC = 0.84, 95% CI = 0.75–0.94), while results for liver-ϕ were poor, at 46% (95% CI = 33–59) and moderate, at 82% (95% CI = 65–100; AUC = 0.60, 95% CI = 0.47–0.73). The inferior diagnostic power of liver-ϕ compared with tumor-ϕ indicates noncorrelated fluid properties of liver and embedded lesions. Liver-c of the healthy volunteers showed a stiffness below the cut-off value (1.39 ± 0.04 m/s). Results of ROC analysis for characterization with c and ϕ are shown in Fig. 3C and Supplementary Table S3.

Liver-c correlated with tumor-c (P < 0.001; r = 0.74), indicating that tumors preferentially grow in stiff livers, while liver-ϕ correlated only moderately with tumor-ϕ (P < 0.01; r = 0.34) in agreement with the aforementioned insensitivity of liver fluidity to fibrosis.

To the best of our knowledge, this is the first study generating fluidity maps for the characterization of liver tumors. We believe that whole-organ tomographic mapping of stiffness and fluidity with high spatial resolution to depict multiple lesions in the same liver, including lesions smaller than 10 millimeters, is important for quantitative biomechanics-based diagnosis of liver tumors.

For further discussion, it is important to note the contributions of elasticity (the tissue's ability to store mechanical energy) and viscosity (its ability to convert mechanical energy into heat) to our parameters. Because c is related to magnitude shear modulus |G*| by c² = 2·|G*|/(ρ·[1+cos ϕ]), it mixes contributions from elasticity and viscosity. Garteiser and colleagues found the loss modulus to be significantly larger in HCCs than in benign hepatocellular tumors, while storage modulus did not differ between these lesion entities (8). Those authors also reported larger |G*| values for malignant than for benign tumors, probably caused by an increase in loss properties. The sensitivity of the loss modulus to tumor malignancy might be explained by an increase in fluidity due to vascularization or the presence of necrotic tissue. Furthermore, malignant tumors can accumulate hydrophobic, disorganized proteins, which increase tissue stiffness and mechanical friction. This could explain why malignant lesions present as stiff, yet fluid, masses with ϕ values above the limit of solid–fluid transitions (π/4 = 0.78 rad). Other mechanisms potentially decrease ϕ as sketched in Supplementary Fig. S5A–S5C. For example, large organized networks such as portal-to-portal bridges in liver fibrosis or polar water–binding ECM components could reduce ϕ. In liver fibrosis, the mechanisms that either drive or inhibit fluidity seem to be balanced, rendering the overall fluid–solid behavior of liver tissue insensitive to the progression of fibrosis (7). Similarly, our parameter ϕ in tumor surrounding liver tissue was not sensitive to malignancy. The apparent insensitivity of ϕ to liver fibrosis makes fluidity a promising biomechanical marker, with a large difference between tumors and surrounding collagen-rich liver tissues regardless of present fibrosis. Furthermore, in our cohort, abnormal fluidity was specific for malignancy (ruling out solid, benign lesions) while tumor stiffness was sensitive to malignancy (detecting stiff, malignant lesions), thus rendering the combination of the two MRE-based parameters useful for tumor characterization.

Although fluidity was greater in malignant than benign tumors, we observed high ϕ values in hemangioma, which can be explained by their vascularity and fluid structure. Vascularity has been reported to increase liver stiffness and viscosity (16). The diagnosis of hemangioma is well established through the European Association for the Study of the Liver guidelines (17). In contrast, hepatic adenoma and FNH can be misinterpreted as malignant tumors when classified by contrast-enhanced MRI alone (18). To address this radiological ambiguity, we performed a second analysis without hemangioma and identified hepatic adenoma and FNH versus malignant masses. In this analysis, overall diagnostic power further increased to 94% sensitivity for c and 92% specificity for ϕ. It should be noted that in 2 patients with large FNH we observed stiffer and more fluid mechanical properties, similar to those of malignant masses. In those cases, presented in Supplementary Fig. S4A and S4B, hypervascularization due to hyperplastic response to portal tract injury (19, 20) may have been responsible for elevated ϕ values, while collagen deposition due to scaring could explain higher c values.

Although encouraging, our study has limitations. First of all, by nature of its design, only patients with known liver lesions who needed further diagnostic MRI workup or treatment were enrolled. On the one hand, the distribution of tumor entities in our cohort is typical for a tertiary-care liver center. On the other hand, patient selection is therefore biased toward malignant lesions, as individuals with benign liver tumors (except hepatic adenoma) rarely undergo treatment. For this reason, our study cohort includes a relatively small number of patients with benign liver tumors. Second, not all patients with benign lesions had histopathologic confirmation, but were diagnosed by established imaging criteria in contrast-enhanced MRI. Similarly, not all patients with stiff livers obtained biopsy for confirmation of liver fibrosis or cirrhosis, precluding a correlation analysis of tomoelastography and fibrosis grade. Finally, tomoelastography requires specialized hardware, making set-up and clinical dissemination of the method more difficult than for other innovative MRI methods. Nevertheless, prospective multicenter studies in larger cohorts are planned.

In conclusion, solid–fluid parameters measured by tomoelastography are a promising source of additional diagnostic information in patients with liver tumors including small lesions below 10 mm in diameter. Our results indicate that changes in liver stiffness are related to changes in solid-tissue properties, while changes in tumor stiffness are related to fluid-tissue properties. We propose fluidity as a new imaging marker that holds great promise for tumor detection, irrespective of the stage of liver fibrosis; in conjunction with tissue stiffness, it may allow highly sensitive and specific characterization of liver tumors.

T. Denecke reports receiving commercial research support and honoraria from Speakers Bureau from Siemens. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Shahryari, U. Fehrenbach, T. Denecke, I. Sack

Development of methodology: M. Shahryari, J. Braun, I. Sack

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Shahryari, G. Böning, U. Fehrenbach, T. Denecke

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Shahryari, H. Tzschätzsch, J. Guo, S.R. Marticorena Garcia, P. Asbach, J.A. Käs, I. Sack

Writing, review, and/or revision of the manuscript: M. Shahryari, J. Guo, S.R. Marticorena Garcia, G. Böning, U. Fehrenbach, L. Stencel, P. Asbach, B. Hamm, J. Braun, T. Denecke, I. Sack

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Shahryari, S.R. Marticorena Garcia, G. Böning, U. Fehrenbach, L. Stencel, P. Asbach, B. Hamm, J. Braun, T. Denecke, I. Sack

Study supervision: T. Denecke, I. Sack

Support of the German Research Foundation (SFB1340 “Matrix in Vision”, GRK2260 BIOQIC to I. Sack, J. Braun, J. Guo, P. Asbach, and B. Hamm) and the German Federal Ministry of Education and Research (LiSyM 031L0057 to I. Sack) is gratefully acknowledged.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.