Unsupervised Generation of Pseudo Normal PET from MRI with Diffusion Model for Epileptic Focus Localization

TL;DR

This work tackles epileptic focus localization when MRI is inconclusive by generating pseudo normal FDG-PET images from MRI using unsupervised unpaired MR-to-PET translation. It compares a diffusion-based adversarial method (SynDiff) with CycleGAN, finding that SynDiff produces higher-fidelity pseudo normal PET and improves localization accuracy. Localization relies on a Z-score analysis of the difference between real PET and pseudo normal PET, with $\text{Z} = (X - \mu)/\sigma$ and a threshold of $\text{Z} < -1.65$ (P = 0.05) plus a cluster size constraint. The results suggest that diffusion-based image synthesis can robustly provide personalized control PET in the absence of matched healthy data, potentially enhancing surgical planning for epilepsy.

Abstract

[$^{18}$F]fluorodeoxyglucose (FDG) positron emission tomography (PET) has emerged as a crucial tool in identifying the epileptic focus, especially in cases where magnetic resonance imaging (MRI) diagnosis yields indeterminate results. FDG PET can provide the metabolic information of glucose and help identify abnormal areas that are not easily found through MRI. However, the effectiveness of FDG PET-based assessment and diagnosis depends on the selection of a healthy control group. The healthy control group typically consists of healthy individuals similar to epilepsy patients in terms of age, gender, and other aspects for providing normal FDG PET data, which will be used as a reference for enhancing the accuracy and reliability of the epilepsy diagnosis. However, significant challenges arise when a healthy PET control group is unattainable. Yaakub \emph{et al.} have previously introduced a Pix2PixGAN-based method for MRI to PET translation. This method used paired MRI and FDG PET scans from healthy individuals for training, and produced pseudo normal FDG PET images from patient MRIs that are subsequently used for lesion detection. However, this approach requires a large amount of high-quality, paired MRI and PET images from healthy control subjects, which may not always be available. In this study, we investigated unsupervised learning methods for unpaired MRI to PET translation for generating pseudo normal FDG PET for epileptic focus localization. Two deep learning methods, CycleGAN and SynDiff, were employed, and we found that diffusion-based method achieved improved performance in accurately localizing the epileptic focus.

Figures (8)

• Figure 1: The proposed workflow for synthesizing pseudo normal PET images. It consists of two components: (a) adversarial diffusion model and (b) adversarial projector.
• Figure 2: Training process of SynDiff. During training, each modality consists of (b) diffusive and (c) non-diffusive modules. Here only shows the diffusive and non-diffusive modules corresponding to the target modality $\boldsymbol{x}$ (PET), the process corresponding to the source modality $\boldsymbol{y}$ (MRI) are analogous.
• Figure 3: Illustration of pseudo normal PET assisted epileptogenic focus localization diagnosis workflow.
• Figure 4: Results of MRI-to-PET image translation. Anatomical MR images and its corresponding real FDG PET image are shown on the left. Generated PET images by CycleGAN and SynDiff, as well as their corresponding error maps are displayed side-by-side.
• Figure 5: Singular value spectrum comparison between SynDiff- and CycleGAN-generated PET images and real PET images.