AUTOMATED SEGMENTATION OF CORPUS CALLOSUM IN BRAIN MR IMAGES IN ALZHEIMER’S CONDITIONS USING IMPROVED UNET++ MODEL

Abstract

The Corpus Callosum (CC) is a large white matter bundle that connects the left and right cerebral hemispheres of the human brain. It is susceptible to atrophy as Alzheimer’s disease progresses. The robust segmentation of CC allows quantitative investigation of its structural changes. However, deep learning-based CC segmentation is less explored. In this work, an improved UNet model is proposed for CC segmentation from two-dimensional T1-weighted mid-sagittal brain MRI. For this, mid-sagittal scans (n = 184) from the publicly available Open Access Series of Imaging Studies (OASIS) brain MRI database are used. The images are fed to an improved UNet++ network. The architecture contains a fully convolutional network with two paths, contracting and extracting, that are connected in a U-shape to automatically extract spatial information. Leave one out Cross-Validation (LooCV) method is used to evaluate the robustness of the proposed method. Results show that the proposed approach is able to segment CC from MR images. The proposed method yields the Dice score of 98.43%, and Jaccard index of 98.53%. The improved UNet++ model obtained the highest sensitivity of 99.21% for AD conditions. Further, the performance of the proposed model has been validated against the state-of-the-art methods. Thus, the proposed approach could be useful for the segmentation of MR images in clinical condition.