Segmentation of white matter hyperintensities on ${}^{18}$F-FDG PET/CT images with a generative adversarial network

White matter hyperintensities (WMH) are typically segmented using MRI because WMH are hardly visible on ${}^{18}$F-FDG PET/CT. This retrospective study was conducted to segment WMH and estimate their volumes from ${}^{18}$F-FDG PET with a generative adversarial network (W${}_{hyper}$GAN). We selected patients whose interval between MRI and FDG PET/CT scans was within 3 months, from January 2017 to December 2018, and classified them into mild, moderate, and severe groups by following the semiquantitative rating method of Fazekas. For each group, 50 patients were selected, and of them, we randomly selected 35 patients for training and 15 for testing. WMH were automatically segmented from FLAIR MRI with manual adjustment. Patches of WMH were extracted from ${}^{18}$F-FDG PET and segmented MRI. W${}_{hyper}$GAN was compared with H-DenseUnet, a deep learning method widely used for segmentation tasks, for segmentation performance based on the dice similarity coefficient (DSC), recall, and average volume differences (AVD). For volume estimation, the predicted WMH volumes from PET were compared with ground truth volumes. The DSC values were associated with WMH volumes on MRI. For volumes >60 mL, the DSC values were 0.751 for W${}_{hyper}$GAN and 0.564 for H-DenseUnet. For volumes $\le$60 mL, the DSC values rapidly decreased as the volume decreased (0.362 for W${}_{hyper}$GAN vs. 0.237 for H-DenseUnet). For recall, W${}_{hyper}$GAN achieved the highest value in the severe group (0.579 for W${}_{hyper}$GAN vs. 0.509 for H-DenseUnet). For AVD, W${}_{hyper}$GAN achieved the lowest score in the severe group (0.494 for W${}_{hyper}$GAN vs. 0.941 for H-DenseUnet). For the WMH volume estimation, W${}_{hyper}$GAN performed better than H-DenseUnet and yielded excellent correlation coefficients (r = 0.998, 0.983, and 0.908 in the severe, moderate, and mild group). Although limited by visual analysis, the W${}_{hyper}$GAN based can be used to automatically segment and estimate volumes of WMH from ${}^{18}$F-FDG PET/CT. This would increase the usefulness of ${}^{18}$F-FDG PET/CT for the evaluation of WMH in patients with cognitive impairment.