Quantifying the Detected Contrast as a Function of Lesion Size in Ultrafast Breast MRI

Abstract

Problem: Malignant breast lesions tend to rapidly uptake and washout contrast and conventional MRI sequences do not image fast enough to capture these enhancement kinetics. The consequence is poor specificity of breast MRI in distinguishing benign from malignant lesions, resulting in biopsies. However, imaging with ultrafast protocols can result in a tradeoff between temporal and spatial resolution. Objective: The objective of this thesis was to determine the optimal spatial resolution of an ultrafast breast MRI protocol to maximize contrast and contrast to noise ratio (CNR) of ~ 5 mm lesions for use in clinical breast MR imaging. Methods: We used a combination of in-silico simulations, 3D-printed phantoms and clinical data to analyze the effect that voxel size has on the contrast and CNR of various lesion sizes in the MR images. We analyzed isotropic voxels, non-isotropic voxels, and ultrafast (TWIST) acquisitions. The lesion size were 50% of the contrast or CNR was lost was identified for each acquisition and compared to both the nominal voxel size and the target clinical lesion size of ~5 mm. Additionally, we compared the spatial resolution of each acquisition by calculating the FWHM of the line spread function and comparing it to the cutoff frequency. For our clinical data analysis, we measured the cross-sectional diameter of patient vasculature of a high-resolution MRI and the CNR on the TWIST image to calculate the CNR for a given lesion size. This was done in three and then compared to the 3D printed CNR TWIST data. Results: Our simulated and 3D printed data shows that the lesion size where 50% of contrast and CNR is lost is near the nominal resolution. The measured FWHM decreased with increasing cutoff frequency for our isotropic voxel images. Our results demonstrate that for a fixed voxel volume, an isotropic voxel may result in higher CNR and contrast for clinically relevant lesion sizes when compared to non-isotropic voxels. The clinical data for each patient showed a trend where the CNR decreased for decreasing lesion size and showed agreement with our 3D printed phantom results. The FWHM values for the TWIST acquisitions suggest that the selected A and B fractions have very little impact on the spatial resolution. Our phantom results suggest that isotropic voxel sizes between 2.5 and 3 mm have higher CNR than the current clinical protocol of 1.56 x 1.56 x 1.25 mm for a lesion size of ~ 5 mm. Voxel sizes in this range still have ~85% of their CNR. Discussion: The loss of contrast and CNR in our data was due to the partial volume effect. The agreement between our 3D printed data and our clinical data suggests that our phantom is an effective model at demonstrating how contrast and CNR degrade with decreasing lesion sizes. In the context of breast MRI, increasing the voxel size could allow for faster scan times. Our spatial resolution results suggest that calculating the FWHM of the line spread function through an edge on an MRI, may be an effective metric to compare the spatial resolutions of different acquisitions. Our TWIST results showed that the A and B fractions are not the determining factors in the spatial resolution of the acquisition. However, our preliminary results suggest that choosing a low A and B fraction may reduce the CNR of the acquisition. Our data suggests that using isotropic voxels between 2.5 and 3 mm may allow for improved detection of enhancing lesions. Future work will investigate CNR of enhancing lesions acquired with isotropic voxel sizes between 2.5 and 3 mm. Conclusions: While these initial findings serve as useful proof-of-concept, future studies should aim to further validate these results in larger patient cohort. Furthermore, extending the analysis to include temporal dynamics would provide a more complete representation of ultrafast capabilities.