Plasma-CycleGAN: Plasma Biomarker-Guided MRI to PET Cross-modality Translation Using Conditional CycleGAN

TL;DR

This work addresses the challenge of translating MRI to PET images for Alzheimer's disease by leveraging plasma biomarkers to condition the generation process. By evaluating three baseline cross-modality models and integrating the plasma $A\beta_{42}/A\beta_{40}$ ratio in three ways, the authors demonstrate consistent improvements in image quality and amyloid-related metrics, with Plasma-CycleGAN delivering the best performance. The findings show stronger correlations and higher amyloid-positivity classification accuracy for synthesized PET images when BBBMs are included, indicating potential reductions in PET usage and radiation exposure in clinical settings. The study highlights the feasibility and value of incorporating blood-based biomarkers into conditional cross-modality synthesis and lays groundwork for extending to more advanced models and additional biomarkers such as $p$-tau.

Abstract

Cross-modality translation between MRI and PET imaging is challenging due to the distinct mechanisms underlying these modalities. Blood-based biomarkers (BBBMs) are revolutionizing Alzheimer's disease (AD) detection by identifying patients and quantifying brain amyloid levels. However, the potential of BBBMs to enhance PET image synthesis remains unexplored. In this paper, we performed a thorough study on the effect of incorporating BBBM into deep generative models. By evaluating three widely used cross-modality translation models, we found that BBBMs integration consistently enhances the generative quality across all models. By visual inspection of the generated results, we observed that PET images generated by CycleGAN exhibit the best visual fidelity. Based on these findings, we propose Plasma-CycleGAN, a novel generative model based on CycleGAN, to synthesize PET images from MRI using BBBMs as conditions. This is the first approach to integrate BBBMs in conditional cross-modality translation between MRI and PET.

Figures (2)

• Figure 1: Our proposed architecture for BBBM incorporation (using CycleGAN as an example). We implemented three ways to introduce plasma A$\beta$42/40 values into generative frameworks: a) normalized plasma A$\beta$42/40 levels were expanded to the input image size and added to the image. b) normalized plasma A$\beta$42/40 levels were expanded to the size of the latent feature map and added to the feature map. c) normalized plasma A$\beta$42/40 levels were expanded to the size of one channel in latent feature map and concatenated in latent space. $D_1$ and $D_2$ are two discriminators. Pix2pix and ShareGAN have similar architectures, and the BBBM information was added using the same method.
• Figure 2: Visualization of generate PET images. By visual inspection, CycleGAN generated PET images with more reliable pixel-wise distribution than Pix2pix and ShareGAN. The three rows correspond to axial, sagittal and coronal views, respectively.