Abstract
There is an increasing need of multi-channel fluorescent imaging to simultaneously visualize different biological structures and the dynamic interactions between them. However, due to the overlapping emission spectra of many commonly used fluorophores, many times fluorescent images have bleed-through problems when a fluorophore in one channel crosstalk with another channel, leading to a false positive detection. This problem is almost unavoidable when the samples are stained with three or more dyes, and especially imaged under multiphoton microscope in which case the fluorophores two-photon excitation spectra are usually broader than one-photon. A lot of methods have been used to do spectral unmixing, such as non-negative matrix factorization and similarity unmixing. However, they either need known emission spectra or assume one fluorophore per pixel.
To help users solve channel bleed-through problem, the Multiphoton Imaging (MP) core at URMC developed a novel method by using K-means clustering to separate mixed channels and clear up images. With a given multiphoton image, pixels are classified into K different clusters based on their intensities in all the detection channels, with a minimal within-cluster sum of squares. Pixels bleed into another channel will be separated out from that channel's cluster due to a different intensity signature. Surprisingly, this method also works very well to remove background noise and autofluorescence by treating the background or autofluorescence signal as an additional cluster. Furthermore, as a core, we integrated this tool into Imaris for users without any programming knowledge to use it easily.
This newly developed tool for image pre-processing has been used by many users at the MP Core and is going to be applied for more image modalities such as confocal. The success of this development has initiated our image analysis service, and we are continuing building an Image Analysis Platform with the focus of more customized image analysis solutions.