Toward deep MRI segmentation for Alzheimer’s disease detection

Alzheimer’s disease (AD) is an irreversible, progressive, and ultimately fatal brain degenerative disorder, no effective cures for it till now. Despite that, the available treatments can delay its progress. So, early detection of AD plays a crucial role in preventing and controlling its progress. Hippocampus (HC) is among the first impacted brain regions by AD. Its shape and volume are measured using a structural magnetic resonance image (MRI) to help AD diagnosis. Therefore, brain hippocampus segmentation is the building block for AD detection. This study’s main objective is to propose a deep learning Alzheimer’s disease hippocampus segmentation framework (DL-AHS) for automatic left and right hippocampus segmentation to detect and identify AD. The proposed DL-AHS framework is based on the U-Net architecture and estimated on the baseline coronal T1-weighted structural MRI data obtained from Alzheimer’s disease neuroimaging initiative (ADNI) and neuroimaging tools and resources collaboratory (NITRIC) datasets. The dataset is processed using the Medical Image Processing, Analysis, and Visualization (MIPAV) program. Besides, it is augmented using a deep convolutional generative adversarial network (DC-GAN). For left and right HC segmentation from other brain sub-regions, two architectures are proposed. The first utilizes simple hyperparameters tuning in the U-Net (SHPT-Net). The second employs a transfer learning technique in which the ResNet blocks are used in the U-Net (RESU-Net). The empirical results confirmed that the proposed framework achieves high performance, 94.34% accuracy, and 93.5% Dice similarity coefficient for SHPT-Net. Also, 97% accuracy and 94% Dice similarity coefficient are achieved for RESU-Net.