Automatic Hippocampus Volume Estimation Using 3D UNet for Alzheimer Detection

The detection of Alzheimer’s disease at early stage is essential for deploying interventions and for the disease retrogression. Hippocampus volume atrophy is a definitive indicator of the early onset of Alzheimer's. Measuring hippocampus volume from 3D images manually is tedious and prone to errors. There exist algorithms based on statistics and segmentation to estimate the hippocampus volume from the 3D medical images automatically. These methods suffer from inter-individual variability and thus cannot be used out of the box for all patients. Thus, we propose an algorithm for automatic hippocampus volume estimation. The proposed model can be successfully deployed for patients across age and gender. The proposed deep learning model is trained using Adam optimizer on the medical decathlon dataset. The model estimates the hippocampus region in the given 3D MRI image and its volume. The estimated volume is compared with the manually labelled volume using Dice and Jaccard similarity scores.