Deep Learning-Based Imaging Technology FLI-Net

2019-11-16

On November 13, 2019, researchers at Rensselaer Polytechnic Institute developed a new deep neural network technology called FLI-Net, which applies deep learning to quantify images generated by fluorescence lifetime imaging. Molecular imaging has long been an effective tool for researchers to understand the fundamental biological components that may lead to clinical applications. Specifically, fluorescence lifetime imaging (FLI) provides a non-invasive method to examine samples and gain unique insights into the cellular microenvironment. FLI can be used to quantify interactions between proteins in vivo, as well as the binding of biosensor ligands to receptors. However, this indirect imaging method requires iterative optimization and post-processing of collected data, which greatly limits the system. Traditionally, analyzing this data requires a significant amount of time and complex mathematical tools that heavily rely on the user, making it difficult to generate consistent and reproducible images. These challenges have been obstacles to the use of this technology in clinical settings. The research team developed a 3D convolutional neural network (CNN) specifically designed to process data from FLI systems. This system does not require any user-defined input parameters. Furthermore, a synthetic data generator can effectively train the system without the need for a large training dataset obtained through experiments. The CNN can handle experimental fluorescence decay obtained through time-correlated single-photon counting (TCSPC) or gated ICCD instruments (the most common FLI technique). The 3D CNN, called "FLI-Net," aims to simulate curve fitting methods using convolutional operations and nonlinear activation function layers, allowing for input of fluorescence decay with time and spatial resolution in a 3D data cube (x, y, t), and estimating bi-exponential parameters (two lifetimes and one fractional amplitude) at each pixel to provide an output image of the same size as the input (x, y) dimensions. To test FLI-Net, researchers used FLI to image cancer cells and visualize the metabolic state of living cells, as well as measure the level of receptor involvement using Förster resonance energy transfer (FRET). The research group found that the performance of the deep neural network was comparable to commercially available software and even superior in certain cases. Additionally, they discovered that this technology requires less light to generate detailed images. Xavier Intes, a professor of biomedical engineering at Rensselaer, who led the study, said, "The tools we provide will be more suitable for end users, not only biologists but also surgeons." The researchers found that the FLI-Net architecture is well-suited for imaging formation patterns. Therefore, their deep learning framework can serve as a universal new tool for non-fitting analysis of complex fluorescence lifetime imaging processes. Pingkun Pan, an assistant professor of biomedical engineering at Rensselaer, said, "The goal of this research is to bring the benefits of FLI as a key tool for precision medicine into clinical practice, which is a technology in many clinical applications. For example, it can be used for real-time in vivo imaging of tumors, which can help surgeons visualize lesions during surgical procedures, enabling them to completely remove cancer tissue with minimal damage to healthy tissue."

RECENT AI NEWS

RECENT AI TOOLS