Improved YOLOv8 for early detection of Alzheimer's disease from magnetic resonance imaging.

Background: Alzheimer's disease (AD) is a life-threatening condition affecting 47 million people globally, with 13% of them over the age of 65. Despite ongoing efforts in drug development, the incidence of Alzheimer's disease is increasing, projected to reach 131 million new cases in the next two decades. Delayed diagnosis and failure to uncover the neuropathological pathways of AD contribute to the increasing incidence and sequelae of AD. Methods: Here, we address these challenges by developing a model derived from YOLOv8 for the early diagnosis of AD. We improve YOLOv8 by replacing the CBS convolution modules with the RepVGG (Reparameterized VGG) module and appending the Spatial Pyramid Pooling Enhanced with ELAN (SPPELAN) module and Simple Attention Module (SimAM) at the end of the YOLOv8 network. Additionally, we improve the model by introducing C2f_EMA module at YOLOv8 neck. Results: Our results demonstrate a significant improvement in our model's performance compared to the benchmark YOLOv8 for AD detection. While the benchmark YOLOv8 showed performance metrics of 0.794 accuracy, 0.876 recall, 0.881 mAP50, and 0.875 mAP50:95, our model demonstrated improved performance with 0.816 accuracy, 0.877 recall, 0.904 mAP50, and 0.898 mAP50:95, indicating improvements of 2.77, 0.11, 2.61, and 2.62%, respectively. These findings provide insights into the possibility of achieving an effective diagnosis at the early stage of AD, which may aid early personalized intervention and further assist researchers in conducting in-depth examinations to understand the early mechanisms underlying AD.