Alzheimer's disease detection using Multi Atlas Segmentation based on Temporal Group Sparse Regression

• Published in: 2023 International Conference on New Frontiers in Communication, Automation, Management and Security (ICCAMS) Vol. 1; pp. 1 - 13
• Main Authors: Velkumar, K, Chokkalingam, Bala Subramanian, Solairaj, Dr A, Vignesh, L.S
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.10.2023
• Subjects: Correlation; Data models; Feature extraction; Genomics; HSVM; Imaging; Neuroimaging; Support vector machines; T-GSRAM
• DOI: 10.1109/ICCAMS60113.2023.10526013

Imaging genomics stands as a valuable approach for identifying potential genetic factors associated with Alzheimer's disease (AD) within genomic and imaging datasets. Most existing techniques in imaging genomics employ a linear model to investigate the connection between quantitative traits (QTs) derived from brain imaging and genetic data (single nucleotide polymorphisms). This method often fails to consider the connections between clusters of Quantitative Traits (QTs) and Single Nucleotide Polymorphisms (SNPs), as well as the complex interactions which involves longitudinal imaging quantitative traits and Single Nucleotide Polymorphisms. In order to tackle these constraints, we introduce a fresh approach named Temporal Group Sparsity Regression and Additive Model (TGSRAM). This pioneering technique strives to reveal associations between longitudinal imaging QTs and SNPs, with the objective of pinpointing potential biomarkers for Alzheimer's Disease (AD). Our approach seeks to offer solutions to these challenges. Furthermore, we introduce an alternative methodology termed multi-atlas-based morphometry in our study. This methodology assesses morphometric representations of the same image across numerous atlas spaces. By generating representations from various atlases, we harness supplementary insights to distinguish between different groups and mitigate the adverse impacts stemming from registration errors. In practice, each subject under investigation undergoes registration to multiple atlases, from which adaptive regional features are extracted. Subsequently, a correlation and relevance-based approach is employed to collectively select features from diverse atlases. This is followed by the ultimate classification step using the hybrid support vector machine (HSVM).