Morphological feature visualization of Alzheimer's disease via Multidirectional Perception GAN

The diagnosis of early stages of Alzheimer's disease (AD) is essential for\ntimely treatment to slow further deterioration. Visualizing the morphological\nfeatures for the early stages of AD is of great clinical value. In this work, a\nnovel Multidirectional Perception Generative Adversarial Network (MP-GAN) is\nproposed to visualize the morphological features indicating the severity of AD\nfor patients of different stages. Specifically, by introducing a novel\nmultidirectional mapping mechanism into the model, the proposed MP-GAN can\ncapture the salient global features efficiently. Thus, by utilizing the\nclass-discriminative map from the generator, the proposed model can clearly\ndelineate the subtle lesions via MR image transformations between the source\ndomain and the pre-defined target domain. Besides, by integrating the\nadversarial loss, classification loss, cycle consistency loss and \\emph{L}1\npenalty, a single generator in MP-GAN can learn the class-discriminative maps\nfor multiple-classes. Extensive experimental results on Alzheimer's Disease\nNeuroimaging Initiative (ADNI) dataset demonstrate that MP-GAN achieves\nsuperior performance compared with the existing methods. The lesions visualized\nby MP-GAN are also consistent with what clinicians observe.\n