Modelling Vascularity in Breast Cancer and Surrounding Stroma Using Diffusion MRI and Intravoxel Incoherent Motion

Colleen Bailey, Sarah Vinnicombe, Eleftheria Panagiotaki, Shelley A. Waugh, John H. Hipwell, Daniel C. Alexander, Kathryn Kitching, Patsy J. Whelehan, Sarah E. Pinder, Andrew Evans, David J. HawkesDigital Mammography / IWDM 2014 · DOI

TL;DR

This paper introduces a novel application of intravoxel incoherent motion (IVIM) diffusion MRI to better characterize the vascularity of breast cancer stroma. By modeling the diffusion signal with a biexponential IVIM model, researchers can distinguish between vascular perfusion and tissue diffusion, offering a more sensitive measure of tumor-associated vasculature than conventional diffusion MRI. This improved characterization could lead to better prediction of tumor invasiveness and recurrence.

What Problem Does This Paper Address?

Predicting breast cancer invasiveness and recurrence is crucial for effective treatment. Changes in the stroma surrounding the tumor are known to be important markers, with enhanced signal in Dynamic Contrast-Enhanced MRI (DCE-MRI) correlating with recurrence and survival. However, DCE-MRI's signal is influenced by many physiological parameters, making it complex to interpret. Conventional diffusion MRI, while sensitive to tissue microstructure, has shown limited ability to detect the subtle vascular changes in the stroma that are known to exist and decrease with distance from the tumor. This gap in sensitivity means that important information about the tumor microenvironment might be missed. This study aims to address this by using a more sophisticated diffusion MRI model to better capture vascular contributions in the peri-tumoral stroma.

Methods

The researchers employed a biexponential intravoxel incoherent motion (IVIM) model to analyze diffusion MRI data. This model separates the diffusion signal into two components: one representing pseudodiffusion from blood flow within the vasculature (characterized by parameters $f_p$ and $D_p$) and another representing diffusion in the extravascular tissue (characterized by $D_t$). This contrasts with the standard monoexponential model, which assumes a single diffusion process and is represented by the apparent diffusion coefficient (ADC).

Key methodological steps included:

  • Data Acquisition: Diffusion MRI data was acquired from three pre-chemotherapy breast cancer patients using a 3T Siemens scanner with multiple b-values. DCE-MRI data was also acquired for tumor segmentation.
  • Image Processing: Motion correction was performed using non-rigid registration. Tumors were identified using a region-growing algorithm on DCE-MRI images and verified by a radiologist.
  • Region of Interest (ROI) Definition: ROIs were defined at varying distances from the tumor (0-4 mm, 4-8 mm, 8-12 mm, and 12-16 mm) to analyze the peri-tumoral stroma.
  • Model Fitting: Both monoexponential and biexponential IVIM models were fitted to the diffusion data. Weighted least squares fitting was used, with specific constraints and fixed parameters for $D_p$ based on prior literature.
  • Model Comparison: The Akaike Information Criterion (AIC) was used to compare the goodness-of-fit between the two models, accounting for model complexity.
  • Statistical Analysis: Confidence intervals for the fitted parameters were calculated using a statistical method that adjusts for parameter uncertainty.

Datasets

The study utilized diffusion-weighted MRI and DCE-MRI data from three pre-chemotherapy breast cancer patients. No new public datasets were introduced.

Key Insights and Ideas

  • IVIM's Superiority in Peri-Tumoral Stroma: The IVIM model provided a significantly better fit to the diffusion data in the peri-tumoral stroma within 8 mm of the tumor compared to the conventional monoexponential model. This suggests that vascular contributions are significant in this region and are not adequately captured by simpler models.
  • Vascular Signal Decreases with Distance: The pseudodiffusion fraction ($f_p$), a key IVIM parameter representing vascularity, decreased with increasing distance from the tumor. This finding aligns with known patterns of microvessel density decreasing away from the tumor.
  • ADC Masking Subtle Changes: The monoexponential ADC showed little variation with distance from the tumor. This indicates that while the vascular contribution diminishes, the overall ADC value remains similar, likely due to compensatory increases in extravascular diffusion. This highlights the limitations of ADC in reflecting subtle microenvironmental changes.
  • Potential for Improved Prognostics: The ability of IVIM to quantify vascular contributions in the stroma opens up possibilities for developing more sensitive imaging biomarkers. These biomarkers could potentially improve the prediction of tumor invasiveness and recurrence, complementing existing DCE-MRI information.

Key Results

  • The IVIM model showed consistently lower and more evenly distributed residuals compared to the monoexponential model in the peri-tumoral stroma within 8 mm of the tumor.
  • The pseudodiffusion fraction ($f_p$) was found to be approximately 0.10 +/- 0.03 in the tumor and decreased with distance from the tumor, reaching lower values in more distal stromal regions.
  • The diffusion coefficient ($D_t$) in the stroma was near that of free water and showed an increasing trend further from the tumor, suggesting less restricted diffusion in these areas.
  • The AIC values were lower for the IVIM model in the tumor and peri-tumoral stroma within 8 mm, indicating a better fit.

Limitations and Future Work

  • Small Patient Cohort: The study was conducted on a small number of patients (n=3), which limits the generalizability of the findings.
  • Fixed $D_p$ Parameter: The pseudodiffusion coefficient ($D_p$) was fixed at a value derived from previous studies. While this is a common practice to improve fitting robustness, it might not perfectly represent the specific vascular properties in this cohort.
  • Model Assumptions: The IVIM model assumes that extravascular tissue can be represented by a single diffusion decay and that water does not move between vasculature and tissue during measurement. These assumptions might be less valid in highly complex tumor microenvironments.

Future work suggested by the authors includes:

  • Validation with Histology: Correlating IVIM findings with histological measures of microvessel density to further validate the model's ability to quantify vascularity.
  • Larger Cohorts and Diverse Tumors: Expanding the study to include a larger and more diverse patient population to confirm the observed trends.
  • Exploring More Complex Models: Investigating more sophisticated diffusion models that can account for restricted and hindered water diffusion by cells, particularly within the tumor itself.
  • Clinical Correlation: Investigating the correlation of IVIM-derived stromal vascularity parameters with clinical outcomes such as tumor recurrence and treatment response.

Summary updated

BibTeX
@inproceedings{bailey2014modelling,
  author = {Colleen Bailey and Sarah Vinnicombe and Eleftheria Panagiotaki and Shelley A. Waugh and John H. Hipwell and Daniel C. Alexander and Kathryn Kitching and Patsy J. Whelehan and Sarah E. Pinder and Andrew Evans and David J. Hawkes},
  title = {{Modelling Vascularity in Breast Cancer and Surrounding Stroma Using Diffusion MRI and Intravoxel Incoherent Motion}},
  booktitle = {Digital Mammography / IWDM},
  year = {2014},
  doi = {10.1007/978-3-319-07887-8\_53},
}
Figure