UNMIX-ME

UNMIX-ME (Unmix Multiple Emissions) unmixes spectrally overlapping fluorophores from hyperspectral fluorescence lifetime imaging (HFLI) data by jointly leveraging spectral and temporal (lifetime) signatures for quantitative fluorophore separation.

Key Features:

• Deep learning-based framework: Applies a deep learning model to perform unmixing using combined spectral and temporal information.
• Joint spectral and lifetime unmixing: Integrates spectral and temporal (lifetime) signatures for quantitative separation of fluorophores.
• In silico HMFLI training: Trained and validated using an in silico model that replicates the data acquisition process of a compressive hyperspectral fluorescent lifetime imaging platform (HMFLI).
• Benchmarking versus LSQ: Benchmarked against conventional least squares quantification (LSQ) using tri- and quadri-exponential simulated samples, demonstrating superior unmixing accuracy.
• Handles spectral overlap and bleed-through: Specifically addresses spectral overlap and bleed-through effects between fluorescent probes.
• Applicable to NIR FRET: Supports quantitative analysis for near-infrared Förster Resonance Energy Transfer (NIR FRET) studies.

Scientific Applications:

• Quantitative HFLI imaging: Unmixes multiple fluorescent species in hyperspectral fluorescence lifetime imaging for quantitative molecular assessment of biotissues.
• NIR FRET studies: Enables quantitative fluorophore assessment in near-infrared Förster Resonance Energy Transfer experiments.
• Preclinical in vitro and in vivo imaging: Supports quantitative fluorophore unmixing in both in vitro and in vivo preclinical imaging contexts.

Methodology:

Uses a deep learning-based framework trained and validated on in silico simulations that replicate the compressive HMFLI acquisition and benchmarked against least squares quantification (LSQ) using tri- and quadri-exponential simulated samples.