AAD-DCE: An Aggregated Multimodal Attention Mechanism for Early and Late Dynamic Contrast Enhanced Prostate MRI Synthesis

TL;DR

The paper tackles gadolinium safety in DCE-MRI for prostate cancer by proposing a synthesis approach that relies on multimodal non-contrast MRI inputs to generate early and late post-contrast images. It introduces AAD-DCE, a GAN framework where an Aggregated Attention Discriminator produces global and local attention maps to guide the generator, with the combined attention modulating the input via $x'=(M_x+1)\times x$. The generator operates on concatenated inputs $x$ consisting of T2W, ADC, and T1 pre-contrasts, producing synthetic DCE outputs; experiments on ProstateX show improved PSNR, SSIM, and MAE over baselines, and ablations confirm the value of ADC and attention ensembling, using a compact model of 19.93M parameters. Overall, the method offers a promising, potentially clinically impactful path toward safer DCE-MRI by enabling accurate synthesis from non-contrast imaging modalities.

Abstract

Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is a medical imaging technique that plays a crucial role in the detailed visualization and identification of tissue perfusion in abnormal lesions and radiological suggestions for biopsy. However, DCE-MRI involves the administration of a Gadolinium based (Gad) contrast agent, which is associated with a risk of toxicity in the body. Previous deep learning approaches that synthesize DCE-MR images employ unimodal non-contrast or low-dose contrast MRI images lacking focus on the local perfusion information within the anatomy of interest. We propose AAD-DCE, a generative adversarial network (GAN) with an aggregated attention discriminator module consisting of global and local discriminators. The discriminators provide a spatial embedded attention map to drive the generator to synthesize early and late response DCE-MRI images. Our method employs multimodal inputs - T2 weighted (T2W), Apparent Diffusion Coefficient (ADC), and T1 pre-contrast for image synthesis. Extensive comparative and ablation studies on the ProstateX dataset show that our model (i) is agnostic to various generator benchmarks and (ii) outperforms other DCE-MRI synthesis approaches with improvement margins of +0.64 dB PSNR, +0.0518 SSIM, -0.015 MAE for early response and +0.1 dB PSNR, +0.0424 SSIM, -0.021 MAE for late response, and (ii) emphasize the importance of attention ensembling. Our code is available at https://github.com/bhartidivya/AAD-DCE.

Figures (4)

• Figure 1: (a) Concept diagram of AAD-DCE. Multimodal non-contrast inputs, with aggregated attention maps, synthesize early and late response DCE-MRI. (b) Previous DCE-MRI synthesis methods use adversarial training without focusing on key anatomical regions. The proposed approach integrates a discriminator-based global and local attention mechanism to enhance the generator's performance.
• Figure 2: (a) AAD-DCE architecture with a generator and an Aggregated Attention Discriminator (AAD) module. (b) AAD with local and global attention discriminators, $D_{LA}$ and $D_{GA}$, respectively, utilizes (c) Attention Discriminator (AD) architecture. (d) Embedded attention map $M_{x}$.
• Figure 3: (a) Visual results of early and late response for the ProstateX dataset with error maps. The yellow bounding box marks the ROI. (b) Ablative study with and without ADC maps. (c) Attention maps from different ensembling methods. Attention embedding enables better focus on suspicious regions.
• Figure 4: Generator architecture with AAD enhances focus on the ROI.