Volumetric Hippocampus Segmentation Using 3D U-Net Based On Transfer Learning

The hippocampus, a crucial component of the human brain, is involved in fundamental cognitive processes such as learning, memory, and spatial navigation. However, it is susceptible to several neuropsychiatric disorders, including epilepsy, Alzheimer's disease, and depression. Utilizing Magnetic Resonance Imaging (MRI) techniques with efficient spatial navigation capabilities is crucial for assessing the physiological condition of the hippocampus. Labeling the hippocampus on MRI images primarily depends on manual methods, which are time-consuming and prone to errors between observers. The issue with MRI image processing lies in its demanding computational requirements and lengthy duration. Furthermore, there is a need for more three-dimensional hippocampal datasets for training deep-learning models, in which 3D labeled medical datasets are often scarce in medical imaging. This paper introduces a 3D U-Net architecture that utilizes a transfer learning model to segment the hippocampus from different pre-trained model scenarios. The results of all test scenarios indicate that the suggested model exhibits an average Dice Score, Intersection over Union (IoU) Score, and Sensitivity exceeding 0.85, 0.75, and 0.80, respectively. The proposed methodology enhances the model's ability to generalize within a shorter timeframe, even when dealing with limited volumetric datasets. These results are achieved through transfer learning, which decreases computational complexity by utilizing pre-learned characteristics from previous tasks.