An Attention Module Enhanced Deep Convolutional Neural Network for Automatic Hippocampus Segmentation Using MRI Images

Alzheimer's disease (AD) is a neurodegenerative disorder linked to hippocampal atrophy, affecting cognitive abilities. Clinically, accurate measurement and abnormal detection of the hippocampus based on MRI images are crucial for the diagnosis of AD. However, it is challenging to segment the hippocampal accurately due to anatomical diversity and the low contrast of brain MRI images. In this study, we proposed an improved U-Net model in both 2D and 3D forms. This model uses attention mechanisms for feature fusion and enhancement, aiming to improve the performance of hippocampus segmentation at the edges and low-contrast regions. The proposed model was evaluated using three diverse metrics on two public datasets. Excellent performance was achieved by the proposed model with a best dice coefficient of 0.9261, a best Jaccard of 0.8624 and a best HD95 of 2.6936. The proposed model has the potential to enhance the accuracy of hippocampal segmentation and provide a computer-aided diagnosis tool for AD detection.