CNN Hippodeep AI volumetry correlated with rs-fMRI networks FC in predicting Alzheimer's disease
Poster No:
1764
Submission Type:
Abstract Submission
Authors:
Nur Shahidatul Nabila Ibrahim1, Subapriya Suppiah2, Buhari Ibrahim3, Nur Hafizah Mohad Azmi3, Vengkatha Priya Seriramulu3, Mazlyfarina Mohamad4, Hakimah Mohamad Sallehuddin3, Rizah Mazzuin Razali5, Nor Harzana Harrun6, Marsyita Hanafi2
Institutions:
1KPJ Healthcare University, Nilai, Negeri Sembilan, 2UNIVERSITI PUTRA MALAYSIA, SERDANG, SELANGOR, 3Universiti Putra Malaysia, Serdang, Selangor, 4Universiti Kebangsaan Malaysia, Kuala Lumpur, Wilayah persekutuan, 5Hospital Kuala Lumpur, Kuala Lumpur, Kuala Lumpur, 6Klinik Kesihatan Pandamaran, Klang, Selangor
First Author:
Co-Author(s):
Introduction:
Minor alterations in the size of the hippocampus can happen during normal and abnormal ageing in the human brain. Evaluating hippocampal volumes by manual or even semi-automated means is a laborious process, necessitating the development of fully automated segmentation techniques that are efficient and consistent across time. Artificial intelligence technology using deep convolutional neural networks (CNN) are now being used as effective algorithms for segmenting images in big longitudinal neuroimaging investigations for Alzheimer's Disease (AD). Nevertheless, in order for these innovative algorithms to have practical use in clinical research, it is imperative to verify their accuracy and repeatability by using various datasets.
Methods:
In this study, we assess the effectiveness of a CNN method called Hippodeep, which was created by Thyreau et al., for segmenting the hippocampus. We conducted a comparative analysis of its segmentation outputs with semi-automated segmentation using Voxel-based morphometry (VBM). This analysis was performed on a sample of 15 healthy controls (HC) and 15 AD participants who underwent structural MRI and resting-state functional MRI as well as neuropsychological test questionnaires.
Results:
Findings of (15 AD and 15 HC) showed reduced functional connectivity for resting-state networks such as Default Mode Network (DMN) and Salience Network (SN). HippoDeep model shows higher correlation in assessing hippocampal volume compared to VBM method. Diagnosis accuracy test using ROC curve shows improved sensitivity 93.33% and specificity of 80.00% and AUC 0.927 while right hippocampal volume with sensitivity 80.00%, specificity of 100.00% and AUC 0.91 using HippoDeep model in detecting AD compared to VBM method and ability of HippoDeep model to predict early detection of AD.
·Figure 1: a) Comparison network difference in HC > AD b) Comparison network difference in AD > HC
·Figure 2: ROC curve of the accuracy of the HippoDeep compared to the VBM model for classifying Alzheimer’s disease based on the measurements of bilateral hippocampal volumes. (a)HippoDeep Left Hippoca
Conclusions:
Decreased functional connectivity in resting-state networks can serve as a biomarker, along with HippoDeep, to predict the early onset of Alzheimer's disease.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Segmentation and Parcellation 2
Novel Imaging Acquisition Methods:
Anatomical MRI
BOLD fMRI
Keywords:
Degenerative Disease
Other - Alzheimer's Disease; artificial intelligence; CNN; resting-state fMRI;
1|2Indicates the priority used for review
Provide references using author date format
Ibrahim, N.S.N, Suppiah, S.,Ibrahim, B.,Mohad Azmi, N. H., Seriramulu, V.P., Mohamad, M.,Hanafi,, M., Sallehuddin, H., Razali R. M.,Harrun, N. H. (2023). Validation of automated hippocampus volume assessment using deep learning convolutional neural networks in patients with Alzheimer’s disease medRxiv 2023.04.11.23288432; doi: https://doi.org/10.1101/2023.04.11.23288432
Islam, J., & Zhang, Y. (2019). Understanding 3D CNN Behavior for Alzheimer's Disease Diagnosis from Brain PET Scan. ArXiv, abs/1912.04563.
Liu, L., Zhao, S., Chen, H., Wang, A.: A new machine learning method for identifying alzheimer’s disease. Simulation Modelling Practice and Theory 99, 102023 (2020)