Abstract
Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is a medicalimaging technique that plays a crucial role in the detailed visualization andidentification of tissue perfusion in abnormal lesions and radiologicalsuggestions for biopsy. However, DCE-MRI involves the administration of aGadolinium based (Gad) contrast agent, which is associated with a risk oftoxicity in the body. Previous deep learning approaches that synthesize DCE-MRimages employ unimodal non-contrast or low-dose contrast MRI images lackingfocus on the local perfusion information within the anatomy of interest. Wepropose AAD-DCE, a generative adversarial network (GAN) with an aggregatedattention discriminator module consisting of global and local discriminators.The discriminators provide a spatial embedded attention map to drive thegenerator to synthesize early and late response DCE-MRI images. Our methodemploys multimodal inputs - T2 weighted (T2W), Apparent Diffusion Coefficient(ADC), and T1 pre-contrast for image synthesis. Extensive comparative andablation studies on the ProstateX dataset show that our model (i) is agnosticto various generator benchmarks and (ii) outperforms other DCE-MRI synthesisapproaches with improvement margins of +0.64 dB PSNR, +0.0518 SSIM, -0.015 MAEfor early response and +0.1 dB PSNR, +0.0424 SSIM, -0.021 MAE for lateresponse, and (ii) emphasize the importance of attention ensembling. Our codeis available at https://github.com/bhartidivya/AAD-DCE.