Q-HIU

Q-HIU performs quantitative unmixing of Raman hyperspectral images to retrieve non-negative spatial concentration maps and spectral profiles for chemical identification in samples.

Key Features:

• Efficient Data Processing: Handles large hyperspectral datasets comprising thousands of pixel spectra across many wavenumbers.
• Singular Value Decomposition with Automatic Divisive Correlation: Filters out spatially and spectrally uncorrelated noise autonomously.
• Bottom Gaussian Fitting: Removes fluorescent background interference using a Bottom Gaussian Fitting algorithm.
• Quantitative Unsupervised/Partially Supervised Non-negative Matrix Factorization: Extracts non-negative spatial concentration maps and spectral profiles without prior knowledge or with partial supervision.
• Superior Accuracy and Speed: Produces more accurate quantification and processes data orders of magnitude faster, validated on artificial and real experimental datasets.
• Low-abundance Component Sensitivity: Detects calcium hydroxyapatite and β-carotene at mean Raman concentrations of 0.09% and 0.04%, respectively, despite noise and fluorescent background contributions of 3% and 94%.

Scientific Applications:

• Analysis of human atherosclerotic aortic tissues: Identifies biochemical markers such as cholesterol/cholesterol ester, triglyceride, and calcium hydroxyapatite crystals, and maps oxidatively modified lipids co-localizing with plaque lesions indicative of inflammation and oxidative damage.
• Detection of minor chemical constituents in diseased tissues: Enables identification of low-abundance components (e.g., calcium hydroxyapatite, β-carotene) within samples exhibiting substantial noise and fluorescent backgrounds.

Methodology:

Computational steps comprise Singular Value Decomposition with Automatic Divisive Correlation for noise filtering, Bottom Gaussian Fitting for fluorescent background subtraction, and quantitative unsupervised/partially supervised Non-negative Matrix Factorization to extract non-negative spatial concentration maps and spectral profiles.