Anomaly Analysis of Alzheimer’s Disease in PET Images Using an Unsupervised Adversarial Deep Learning Model

In this study, the anomaly analysis of Alzheimer’s disease using positron emission tomography (PET) images an unsupervised proposed adversarial model is investigated. The consists three parts: a parallel-network encoder, which comprised convolutional pipeline and dilated that extracts global local features concatenates them, decoder reconstructs input image from obtained feature vector, discriminator distinguishes if real or fake. hypothesis trained with only normal brain images, corresponding construction loss for should be minimal. However, belongs to class designated as not with, then will high. This reflect during score comparison between anomalous image. A multi-case performed major classes Disease Neuroimaging Initiative dataset, disease, mild cognitive impairment, control. base parallel-encoder network shows better classification accuracy than benchmark models, built on parallel outperforms detection models. gave out 96.03% 75.21% in area under curve score, respectively. Additionally, qualitative evaluation done by Fréchet inception distance state-of-the-art points.