Synthesizing Late-Stage Contrast Enhancement in Breast MRI: A Comprehensive Pipeline Leveraging Temporal Contrast Enhancement Dynamics

TL;DR

This work tackles the challenge of reducing breast DCE-MRI acquisition time by synthesizing late-phase images from early-phase data while preserving diagnostic interpretability. It introduces a TI-Loss and TI-norm, integrated into a TI-PAN pipeline, to model and reproduce the temporal contrast-enhancement dynamics. Two TI-curve–based metrics, $\mathcal{CP}_{s}$ and $\mathcal{ED}$, are proposed to quantify clinically relevant fidelity beyond pixel realism, and the method is validated on a public 1.5T/3T Duke dataset, showing improved replication of TI patterns in regions of interest and preservation of overall image quality. The approach demonstrates potential to shorten scan times and improve clinical workflow, pending further clinical validation across diverse acquisition protocols and contrast agents.

Abstract

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is essential for breast cancer diagnosis due to its ability to characterize tissue through contrast agent kinetics. However, traditional DCE-MRI protocols require multiple imaging phases, including early and late post-contrast acquisitions, leading to prolonged scan times, patient discomfort, motion artifacts, high costs, and limited accessibility. To overcome these limitations, this study presents a pipeline for synthesizing late-phase DCE-MRI images from early-phase data, replicating the time-intensity (TI) curve behavior in enhanced regions while maintaining visual fidelity across the entire image. The proposed approach introduces a novel loss function, Time Intensity Loss (TI-loss), leveraging the temporal behavior of contrast agents to guide the training of a generative model. Additionally, a new normalization strategy, TI-norm, preserves the contrast enhancement pattern across multiple image sequences at various timestamps, addressing limitations of conventional normalization methods. Two metrics are proposed to evaluate image quality: the Contrast Agent Pattern Score ($\mathcal{CP}_{s}$), which validates enhancement patterns in annotated regions, and the Average Difference in Enhancement ($\mathcal{ED}$), measuring differences between real and generated enhancements. Using a public DCE-MRI dataset with 1.5T and 3T scanners, the proposed method demonstrates accurate synthesis of late-phase images that outperform existing models in replicating the TI curve's behavior in regions of interest while preserving overall image quality. This advancement shows a potential to optimize DCE-MRI protocols by reducing scanning time without compromising diagnostic accuracy, and bringing generative models closer to practical implementation in clinical scenarios to enhance efficiency in breast cancer imaging.

Figures (4)

• Figure 1: Graphical representation of our proposed pipeline. It is composed of three main stages: Generative model training, late post-contrast generation and model evaluation. For pre-processing, we proposed the Time Intensity normalization (TI-norm) to ensure the prevalence of the CA behavior among sequences. In the train stage, we proposed the Time Intensity Loss (TI-Loss) to leverage the contrast-enhancement behavior to outperform the diagnostic value of the generated images. The generation stage allows the synthesis of post-contrast images, and for the evaluation stage we proposed a set of metrics based on the TI curve.
• Figure 2: Results for TI curve estimation using our proposed pipeline. Figure a) shows the per-patient TI curve, where black lines and blue points correspond to the reference for the TI curve (pre-contrast and early post-contrast), orange points are the real CA response (late) and green points are the generated response. Additionally, the red dashed line shows the error distance between real and generated enhancement. Patients are organized according to their expected CA pattern (persistent, plateau and wash-out) and their early post-contrast enhancement in ascendingly. Figure b) shows the distribution of the enhancement increase/decrease among real and generated unannotated regions. Besides, dots in the tails of the distributions represent the outliers in terms of the change in increase.
• Figure 3: Impact of normalization in the diagnostic information. a) Top row displays the early (input) and late (output) responses, besides the generated images using the TI-PAN model. Bottom row shows the time intensity plots for annotated enhanced ROI, where blue continuous line corresponds to the computed real time-intensity curve in late and after normalization, and orange dashed line is the generated time-intensity curve by the model. b) Shows a comparative plot of image quality (PSNR) and CE difference using the model and different normalization strategies in annotated ROIs and unannotated regions.
• Figure 4: Comparative plots for visual image quality. Heatmaps represents the difference of the real vs the generated images in the same scale. Annotated ROIs are also shown in augmented projection. Best seen zoomed in.