HyperVein: A Hyperspectral Image Dataset for Human Vein Detection.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Precise human vein detection is critical in medical diagnostics.
• HyperSpectral (HS) imaging offers rich spectral information but poses computational challenges due to large data volumes.
• Dimensionality reduction techniques are essential for efficient HS image data processing.

Purpose of the Study:

• To present a novel HS image dataset of human hands for vein detection research.
• To evaluate the efficacy of Principal Component Analysis (PCA), Folded PCA (FPCA), and Ward's Linkage Strategy using Mutual Information (WaLuMI) for HS vein detection.
• To compare the performance of these dimensionality reduction techniques when combined with Support Vector Machine (SVM) classification.

Main Methods:

• A new HS image dataset of 100 subjects' hands (left and right) with diverse skin tones was created and anatomically annotated.
• Dimensionality reduction techniques (PCA, FPCA, WaLuMI) were applied to the HS data.
• Support Vector Machine (SVM) binary classification was used, with optimal parameters determined via a training subset; performance was evaluated on a test subset.

Main Results:

• The FPCA-based method achieved the highest accuracy in vein detection compared to PCA and WaLuMI.
• Post-processing using morphological operators enhanced visualization of classification predictions.
• The study confirms the significant potential of HS imaging for accurate vein detection.

Conclusions:

• Folded PCA (FPCA) is a highly effective dimensionality reduction technique for hyperspectral vein detection.
• The developed HS dataset provides a valuable resource for advancing research in medical imaging and diagnostics.
• HS imaging, coupled with appropriate dimensionality reduction, shows promise for non-invasive vein visualization and analysis.