Deep learning-based AI models can improve the segmentation of white matter in 18F-FDG PET/CT images, according to a new study published in the Journal of Digital Imaging. This helps radiologists with the early diagnosis of neurodegenerative disease.
“18F-FDG PET/CT evaluates cortical or subcortical neuronal metabolic activity of the brain and the assessment of the white matter pathologies depends on anatomical imaging modalities such as MRI,” wrote lead author Kyeong Taek Oh, Yonsei University College of Medicine in South Korea, and colleagues. “The potential values of extracting the white matter from 18F-FDG PET/CT have not been evaluated for the quantitative evaluation of various brain diseases.”
The research team tested multiple AI models for the segmentation of white matter, noting that generative adversarial networks (GANs) achieved the most promising results. The GAN model was based on an existing model (pix2pix) and included two convolution networks. One of those networks served as a generator and the other was a discriminator.
“The generator was trained to convert the 18F-FDG PET/CT image to a segmentation map which was to be indistinguishable from the real segmentation map,” the authors explained. “The discriminator was trained to distinguish the generated segmentation map from the real segmentation map. Through adversarial training of generators and discriminators, the generator generated realistic segmentation maps.”
GAN-produced images were inspected by five observers, who assigned a “segmentation quality score” to sample segmentation maps. For the team’s GAN model, 78% of the segmentation results received “adequate” scores. In terms of precision, the model scored a mean value of 0.821 ± 0.036. Looking specifically at recall, the mean value was 0.814 ± 0.029.
These findings provide specialists with a potential new tool for tracking volume change in a patient’s white matter. Such changes have been associated with aging, psychosis and multiple sclerosis. White matter changes are also present in Alzheimer’s patients with “extensive gray matter atrophy,” yet another reason specialists must be able to read these images.
“The segmentation results of the proposed method showed excellent performance mimicking the ground truth images of MRI compared with several commonly used deep learning methods,” the authors concluded. “Further studies are needed to elucidate the clinical implications of FDG PET/CT based white matter segmentation in brain research.”