CLRGAN: Compressed Learning and Reconstruction Using GAN for Alzheimer’s Disease

Alzheimer’s disease is an irreversible brain disorder that mostly affects elderly people. Compressive Sensing Magnetic Resonance Imaging (CS-MRI) is a widely used non-invasive modality to detect Alzheimer’s disease, aiming to reduce the acquisition time. However, CS-MRI makes both the reconstruction and classification problems more challenging due to the low number of the acquired samples. In this paper, a novel conditional Generative Adversarial Network (cGAN) is proposed to jointly classify the compressively sensed data using compressed learning and reconstruct the original image with high visual quality. The proposed network is called Compressed Learning and Reconstruction GAN (CLRGAN). The sensing network within the CLRGAN is proposed to learn the acquisition mask based on the dataset used for training. The results demonstrate that the proposed CLRGAN achieves very high classification accuracy, reaching 99% for a low sampling rate equal to 0.1, competing with methods that utilize the full sampled data for classification. Moreover, the proposed reconstruction network achieves comparable image quality to state-of-the-art networks with lower trainable parameters.