Hippocampus segmentation in magnetic resonance images of Alzheimer's patients using Deep machine learning
This work addresses the need for automated segmentation to aid early detection and monitoring of Alzheimer's disease in patients, but it is incremental as it applies an existing method to a specific medical dataset.
The study tackled hippocampus segmentation in MRI images of Alzheimer's patients using a U-Net deep learning model, achieving a Dice similarity coefficient of 92.3% and sensitivity of 96.5% on test data.
Background: Alzheimers disease is a progressive neurodegenerative disorder and the main cause of dementia in aging. Hippocampus is prone to changes in the early stages of Alzheimers disease. Detection and observation of the hippocampus changes using magnetic resonance imaging (MRI) before the onset of Alzheimers disease leads to the faster preventive and therapeutic measures. Objective: The aim of this study was the segmentation of the hippocampus in magnetic resonance (MR) images of Alzheimers patients using deep machine learning method. Methods: U-Net architecture of convolutional neural network was proposed to segment the hippocampus in the real MRI data. The MR images of the 100 and 35 patients available in Alzheimers disease Neuroimaging Initiative (ADNI) dataset, was used for the train and test of the model, respectively. The performance of the proposed method was compared with manual segmentation by measuring the similarity metrics. Results: The desired segmentation achieved after 10 iterations. A Dice similarity coefficient (DSC) = 92.3%, sensitivity = 96.5%, positive predicted value (PPV) = 90.4%, and Intersection over Union (IoU) value for the train 92.94 and test 92.93 sets were obtained which are acceptable. Conclusion: The proposed approach is promising and can be extended in the prognosis of Alzheimers disease by the prediction of the hippocampus volume changes in the early stage of the disease.