A Deep Learning Approach for Virtual Contrast Enhancement in Contrast Enhanced Spectral Mammography

TL;DR

This work tackles the challenge of reducing contrast media exposure and radiation dose in Contrast Enhanced Spectral Mammography (CESM) by proposing virtual contrast enhancement (VCE) that synthesizes recombined DES images from LE inputs alone. It evaluates three deep-learning models—Autoencoder, Pix2Pix, and CycleGAN—for LE-to-DES translation, with CycleGAN delivering the best overall image fidelity and qualitative realism. A key contribution is the public CESM@UCBM dataset (1138 DICOM images from 105 patients) enriched with BI-RADS/ACR density data, enabling robust benchmarking. The findings indicate that CycleGAN-generated DES images can be difficult to distinguish from real DES images by radiologists and can support BI-RADS assessments, suggesting a viable path toward contrast-free CESM in clinical practice, subject to broader validation.

Abstract

Contrast Enhanced Spectral Mammography (CESM) is a dual-energy mammographic imaging technique that first needs intravenously administration of an iodinated contrast medium; then, it collects both a low-energy image, comparable to standard mammography, and a high-energy image. The two scans are then combined to get a recombined image showing contrast enhancement. Despite CESM diagnostic advantages for breast cancer diagnosis, the use of contrast medium can cause side effects, and CESM also beams patients with a higher radiation dose compared to standard mammography. To address these limitations this work proposes to use deep generative models for virtual contrast enhancement on CESM, aiming to make the CESM contrast-free as well as to reduce the radiation dose. Our deep networks, consisting of an autoencoder and two Generative Adversarial Networks, the Pix2Pix, and the CycleGAN, generate synthetic recombined images solely from low-energy images. We perform an extensive quantitative and qualitative analysis of the model's performance, also exploiting radiologists' assessments, on a novel CESM dataset that includes 1138 images that, as a further contribution of this work, we make publicly available. The results show that CycleGAN is the most promising deep network to generate synthetic recombined images, highlighting the potential of artificial intelligence techniques for virtual contrast enhancement in this field.

Figures (6)

• Figure 1: Illustration of DES images generation. (a) Standard approach with DES image obtained by injecting iodinated contrast medium into the patient and acquiring both the LE and HE images. (b) VCE approach proposed by our work, with DES images generated directly from the LE images, without the need to inject iodinated contrast medium or acquire HE images.
• Figure 2: From left to right, pairs of LE (above) and DES images (below) from the CESM@UCBM dataset with ACR categories a, b, c, d.
• Figure 3: Schematic representation of the methodology.
• Figure 4: Representative diagrams of Autoencoder, Pix2Pix and CycleGAN translating from LE to DES images. Simbols: $x$ is the input image, $\hat{y}$ is the output image, $y$ is the target image, $G$ and $F$ are the generators, $D$ are the discriminators.
• Figure 5: Four representative cases of LE input images, DES target images, and output images of Autoencoder, Pix2Pix, and CycleGAN (from left to right) varying the ACR category (from top to bottom).