Denoising of Fluorescence Lifetime Imaging Data via Principal Component Analysis

TL;DR

This paper introduces a new method to reduce noise in FLIM data, improving the accuracy of detecting metabolic changes in cancer organoids.

Contribution

The novel contribution is the development of noise-corrected PCA (NC-PCA) for denoising FLIM data.

Findings

NC-PCA reduces uncertainty in FLIM data by up to 5.5-fold compared to conventional methods.

NC-PCA reduces data loss over 50-fold while preserving biologically relevant signals.

NC-PCA reveals multiple metabolic states in patient-derived colorectal cancer organoids.

Abstract

Fluorescence Lifetime Imaging Microscopy (FLIM) quantifies autofluorescence lifetime to assess cellular metabolism, therapeutic efficacy, and disease progression. These dynamic and heterogeneous processes complicate signal analysis. Fit-free analysis methods such as phasor analysis are increasingly used due to limitations of fit-based approaches. However, incorporating photon-counting shot noise often leads to moderate-to-high uncertainty in detecting subtle changes. Common noise-reduction strategies can introduce errors and cause data loss. We developed noise-corrected principal component analysis (NC-PCA), which selectively identifies and removes noise to isolate the signal of interest. We validated NC-PCA by analyzing FLIM images of patient-derived colorectal cancer organoids treated with various therapeutics. First, we show NC-PCA decreases uncertainty by up to 5.5-fold compared to conventional analysis and reduces data loss over 50-fold. Then, using a merged dataset, NC-PCA reveals multiple metabolic states. Overall, NC-PCA offers a powerful, generalizable tool to enhance FLIM analysis and improve detection of biologically relevant metabolic changes.

Figures