Characterization of the Diffusion Signal of Breast Tissues using Multi-exponential Models

Background: Diffusion-weighted (DW-) magnetic resonance imaging (MRI) has demonstrated potential as an exogenous contrast-free diagnostic tool for breast cancer screening. Advanced non-Gaussian models of the DW-MRI signal provide insight into the tissue microstructure. Restriction spectrum imaging (RSI) is a mathematical framework that improves tumor conspicuity by decomposing the DW-MRI signal into separate diffusion components. The number of diffusion components and corresponding apparent diffusion coefficients (ADCs) optimal for RSI are organ-specific and determined empirically. The outputs of RSI are the signal contributions of each separable diffusion component. Purpose: To understand the diffusion-weighted MRI signal of cancerous and healthy breast tissues in the context of RSI. Study Type: Retrospective. Populations: 74 women, from two sites, with pathology-proven breast cancer. Field Strength/Sequence: 3.0T Assessment: The DW-MRI signal was described using a linear combination of a variable number of exponential components. The ADC for each component was estimated across all voxels from control and cancer regions of interest (ROIs), patients and sites. Once ADCs were determined, the signal contributions of each diffusion component were estimated using these fixed ADC values. Conventional ADC (mono-exponential) values were also estimated. Statistical Tests: The relative fitting residual and relative Bayesian information criterion (BIC) were assessed. The signal contributions of each diffusion component were compared using analysis of variance (ANOVA) and post-hoc tests. Results: Estimated ADCs for the bi-exponential model were D1,2=2.0x10-5 and D2,2=2.2x10-3 mm2/s, and D1,3=0, D2,3=1.4x10-3 and D3,3=10.2x10-3 mm2/s for tri-exponential model, which in practice is reduced to a bi-exponential model with an offset, or a three-component model. The relative fitting residuals of conventional ADC, bi-exponential and three-component models in control ROIs were 2.1%, 1.62%, and 1.03%, and 3.3%, 1.0%, and 0.3% for cancer ROIs. BIC was smaller for the three-component model, indicating an improved fitting of breast DW-MRI data compared to the bi-exponential model. Conclusion: Breast DW-MRI signal was best described using a tri-exponential model. The signal contributions of the slower component in bi- or tri-exponential models were larger in tumor lesions. These data may be used as differential features between healthy and malignant breast tissues.