1853

A meta-analysis comparing the diagnostic performance of Diffusion-Weighted Imaging (DWI), Diffusion Tensor Imaging (DTI), and Intra-voxel Incoherent Motion (IVIM) in breast cancer
Gabrielle C Baxter1, Martin J Graves2, Fiona J Gilbert1, and Andrew J Patterson2

1Department of Radiology, University of Cambridge, Cambridge, United Kingdom, 2Department of Radiology, Addenbrooke's Hospital, Cambridge, United Kingdom

Synopsis

The performance of parameters from diffusion-weighted imaging (DWI), diffusion tensor imaging (DTI) and intra-voxel incoherent motion (IVIM) in the differential diagnosis of malignant and benign breast lesions were compared through a meta-analysis. 73 eligible studies were included and pooled estimates of sensitivity, specificity and area under the curve were obtained for each of the model parameters. The highest performing parameters for DTI and IVIM were the prime diffusion coefficient (λ1) and the tissue diffusivity (D), respectively. DWI, DTI and IVIM are diagnostically comparable, though there is a lack of standardisation in methodology for each technique.

Introduction

The apparent diffusion coefficient (ADC), measured using diffusion-weighted imaging (DWI), is increasingly used as a biomarker in the detection and characterisation of breast lesions. Recently, advanced diffusion models have been proposed that attempt to capture more complex aspects of the tumour microenvironment. Diffusion tensor imaging (DTI) is a technique that takes into account the anisotropy and directionality of diffusion and measures orthogonal eigenvectors and their eigenvalues, λ1, λ2, and λ3, from which the mean diffusivity (MD), maximal anisotropy index (λ1- λ3) and fractional anisotropy (FA) can be calculated. Intra-voxel incoherent motion (IVIM) separates the effects of diffusion and perfusion by fitting a bi-exponential model to the decay of signal with b-value. In the IVIM model, tissue diffusivity is described by parameter D, pseudo-diffusion or perfusion is given by D*, and the perfusion fraction f. This study reports a meta-analysis comparing the diagnostic performance of DWI, DTI and IVIM and assesses whether or not more advanced techniques achieve an improvement in diagnostic performance that justifies the increased computational complexity and longer scan time required to acquire the range of b-values or diffusion directions required for each model.

Methods

This study was reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRIMSA) guidelines (1). Databases were searched up to March 2018 for studies in English assessing the diagnostic performance of DWI, DTI and IVIM in the breast. Studies were reviewed according to eligibility and exclusion criteria. Publication bias was assessed using funnel plots and Egger’s test was used to measure asymmetry (2). Heterogeneity between studies was assessed using Cochran’s Q-test and Higgins’ I2 test (3). For each of the parameters (ADC, MD, λ1, λ1-λ3, FA, D, f, and D*), forest plots were constructed for sensitivity and specificity. Pooled summary estimates for sensitivity, specificity and area under the curve (AUC) were obtained for each parameter using a bivariate model (4) and summary receiver operating characteristics (sROC) curves were constructed. Analysis was carried out in R (version 3.1.3, https://www.r-project.org/) using the mada package (version 0.5.8, https://CRAN.Rproject.org/package=mada).

Results

From 73 eligible studies (65 ADC studies, 6 DTI studies, 9 IVIM studies), 6791 lesions (3930 malignant and 2861 benign) were included in the meta-analysis. Publication bias was evident for studies using the ADC and significant asymmetry was measured using Egger’s test (p < 0.0001). Results of pooled analysis and heterogeneity measures are given in Table 1. Substantial heterogeneity was present for all parameters except the perfusion fraction (f), however the low number of studies included in the analysis (n=3) for both λ1 and FA resulted in an undefined I2 value for specificity. The pooled sensitivity/specificity/AUC for ADC was 0.89/0.82/0.92. The highest performing parameter for DTI was the prime diffusion coefficient (λ1) with a pooled sensitivity/specificity/AUC of 0.93/0.90/0.94. The highest performing parameter for IVIM was tissue diffusivity (D) with a pooled sensitivity/specificity/AUC of 0.88/0.79/0.90. Forest plots for sensitivity and specificity are presented for all 3 parameters in Figure 1. sROC curves are presented in Figure 2.

Discussion

This is the first study to systematically compare all relevant diffusion techniques for quantitatively distinguishing benign and malignant breast lesions. The pooled sensitivities and specificities using DWI, IVIM and DTI in this meta-analysis were comparable to pooled sensitivity and of dynamic contrast-enhanced MRI specificity (93% and 71%, respectively)(5). For patients with allergies or impaired renal function where the use of a gadolinium-based contrast agent should be avoided, DWI can offer a non-contrast alternative. While IVIM is increasingly used in the breast and achieves a high diagnostic accuracy, there is still a lack of consistent methodology. There is a large variation in the number and range of b-values used and in the choice of parameters reported, with studies often using a combined thresholds approach. DTI also lacks standardisation in methodology and reporting of parameters. The prime diffusion direction λ1 and the mean diffusivity D achieved a diagnostic accuracy equal to or greater than the ADC, however the number of eligible studies included is very low.

Conclusion

DWI, DTI and IVIM are able to discriminate between malignant and benign lesions with a high sensitivity and specificity. IVIM is diagnostically comparable to ADC. DTI achieves a higher accuracy than ADC though the number of studies to date is limited. IVIM and DTI lack standardisation in methodology and parameters reported.

Acknowledgements

No acknowledgement found.

References

1. Moher D, Liberati A, Tetzlaff J, Altman D, The, Group: PRISMA 2009 Checklist. PLoS Med 2009; 6.

2. Egger M, Davey Smith G, Schneider M, Minder C: Bias in meta-analysis detected by a simple, graphical test. BMJ 1997; 315:629–34.

3. Higgins JPT, Thompson SG, Deeks JJ, Altman DG: Measuring inconsistency in meta-analyses. BMJ 2003; 327:557–60.

4. Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH: Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. J Clin Epidemiol 2005; 58:982–990.

5. Zhang L, Tang M, Min Z, Lu J, Lei X, Zhang X: Accuracy of combined dynamic contrast-enhanced magnetic resonance imaging and diffusion-weighted imaging for breast cancer detection: a meta-analysis. Acta radiol 2016; 57:651–660.

Figures

Table 1. Results of pooled estimates and heterogeneity measures

Figure 1. Forest plots of sensitivity and specificity with 95% confidence intervals using apparent diffusion coefficient (ADC), tissue diffusivity (D) and prime diffusion coefficient (λ1). Vertical lines denote pooled summary estimates of sensitivity and specificity.

Figure 2. Summary receiver operating characteristics (sROC) curves for apparent diffusion coefficient (ADC), tissue diffusivity (D) and prime diffusion coefficient (λ1) using the bivariate model with 95% confidence regions.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
1853