A comparison of nonlinear dimensionality reduction algorithms applied on Alzheimer MRI

Sergiu-Adrian Folta, Eva Kaslik

Springer Nature

doi:10.1007/978-3-032-16532-9_24 MRI Machine Learning Alzheimer

Abstract

The study presented in this paper focuses on comparing nonlinear dimensionality reduction techniques applied on an MRI dataset for Alzheimer's disease classification. Magnetic resonance imaging is a critical tool for observing the structural changes in the brain associated with Alzheimer's. However, the high-dimensional nature of MRI data poses significant challenges for effective analysis. Linear dimensionality reduction methods are often inadequate for capturing the complex, nonlinear relationships in the data. Therefore, our study explores the application of NLDR algorithms such as isometric mapping, locally Linear embedding, diffusion maps and t-distributed stochastic neighbor embedding to address these challenges. We used common metrics for clustering performance and trained a classification model to determine the impact on classification. The results indicate that the application of these methods can describe the intrinsic geometric relationships between data points and increase the classification metrics, the best results being obtained by the t-SNE technique.

BibTeX

@inproceedings{Folta_Kaslik_Springer_2026,
  title         = {
    A Comparison of Nonlinear Dimensionality Reduction Algorithms Applied
    on Alzheimer MRI
  },
  author        = {Sergiu-Adrian Folta and Eva Kaslik},
  year          = 2026,
  booktitle     = {
    Advances in Nonlinear Dynamics: Proceedings of the Fourth International
    Nonlinear Dynamics Conference (NODYCON 2025)
  },
  doi           = {10.1007/978-3-032-16532-9_24},
  publisher     = {Springer Nature Switzerland},
  pages         = {271–281},
  isbn          = {978-3-032-16532-9},
  editor        = {Lacarbonara, Walter},
  abstract      = {
    The study presented in this paper focuses on comparing nonlinear
    dimensionality reduction techniques applied on an MRI dataset for
    Alzheimer's disease classification. Magnetic resonance imaging is a
    critical tool for observing the structural changes in the brain associated
    with Alzheimer's. However, the high-dimensional nature of MRI data poses
    significant challenges for effective analysis. Linear dimensionality
    reduction methods are often inadequate for capturing the complex, nonlinear
    relationships in the data. Therefore, our study explores the application of
    NLDR algorithms such as isometric mapping, locally Linear embedding,
    diffusion maps and t-distributed stochastic neighbor embedding to address
    these challenges. We used common metrics for clustering performance and
    trained a classification model to determine the impact on classification.
    The results indicate that the application of these methods can describe the
    intrinsic geometric relationships between data points and increase the
    classification metrics, the best results being obtained by the t-SNE
    technique.
  },
}