Self-Organizing Map and Relational Perspective Mapping for the Accurate Visualization of High-Dimensional...

Area of Science:

• Data Visualization
• Chemometrics
• Mass Spectrometry Imaging

Background:

• Toroidal Self-Organizing Maps (SOMs) offer unsupervised visualization of hyperspectral data by mapping similar spectra to similar colors.
• Previous work demonstrated toroidal SOMs for Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) imaging data.
• Accurate visualization is crucial for interpreting complex hyperspectral datasets.

Purpose of the Study:

• To enhance hyperspectral data visualization accuracy by combining toroidal SOMs with Relational Perspective Mapping (RPM).
• To compare the performance of the novel SOM-RPM technique against established methods like t-distributed stochastic neighborhood embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP).
• To explore the utility of SOM-RPM for data characterization and subsequent analysis using linear discriminant analysis.

Main Methods:

• Optimization of the toroidal Self-Organizing Map (SOM) algorithm.
• Integration of Relational Perspective Mapping (RPM), a nonlinear dimensionality reduction technique, with SOM output.
• Application and comparison of SOM-RPM, t-SNE, and UMAP to Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) imaging data from mouse tumor tissue.

Main Results:

• The combined SOM-RPM approach generates similarity maps that more accurately reflect local spectral distances within hyperspectral data.
• SOM-RPM demonstrates highly competitive performance against t-SNE and UMAP in both qualitative and quantitative evaluations.
• The neural network basis of SOM offers advantages in data characterization.

Conclusions:

• The SOM-RPM technique provides a powerful and accurate method for visualizing and analyzing hyperspectral data, particularly in mass spectrometry imaging.
• SOM-RPM is a valuable alternative to existing visualization methods, offering comparable or superior performance.
• The workflow facilitates further spectral analysis and surface chemistry characterization through linear discriminant analysis.