Evaluation of deep learning models for segmentation of hippocampus volumes from MRI images in Alzheimer’s disease

The hippocampus is a crucial brain structure associated with Alzheimer’s disease (AD). Precise segmentation is crucial for studying AD progression using deep learning. This study aimed to evaluate the performance of deep learning models in segmenting the left and right hippocampus in MRI images. We hypothesized that deep learning-based approaches would enable precise and accurate segmentation of the left and right hippocampus. We propose U-Net, You Only Look Once version 8 (YOLO-v8), and DeepLab-v3 models using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset. This study used 300 subjects, comprising 100 subjects (AD), 100 subjects with mild cognitive impairment (MCI), and 100 subjects with normal control (NC), resulting in a total of 7859 image slices. The results showed that the U-Net model exhibited the best Intersection over Union (IoU), which served as a key performance indicator among the three classes: AD (0.639), MCI (0.801), and NC (0.751). In contrast, YOLO-v8 demonstrated lower IoU performance for AD (0.342), MCI (0.465), and NC (0.550), which are considered inappropriate models to segment the left and right hippocampus. We obtained the left hippocampus volume of AD (1557.5 mm³), MCI (1863.3 mm 3 ), and NC (2089.2 mm 3 ). The right hippocampus volumes of AD (1593.4 mm³), MCI (1918.7 mm 3 ), and NC (2280.2 mm 3 ). The U-Net model exhibited the best performance. We expect deep learning-based methods to assist in clinical decisions by providing accurate hippocampus segmentation.