Deep learning technique replaces fluorescent labeling in cell imaging

Fluorescence microscopy, a popular approach to cell imaging, offers very high biochemical specificity. The technique requires the manipulation or staining of cells with fluorescent labels to best extract proteins when imaging. Of course, it poses challenges. For instance, labeling impacts the basic structure of the cells, often resulting in several hours of delay before the cells can be observed. It also prompts photobleaching and phototoxicity.

Using a laser technique known as 3D holographic microscopy or holotomography to measure the 3D refractive index tomogram of microscopic biological cells and tissues, a team from the Korea Advanced Institute of Science and Technology (KAIST), in conjunction with Tomocube, has demonstrated the quantitative imaging of live cells in real time without using staining or labeling. In addition to quickly and accurately measuring the morphological and structural information of cells, holotomography provides limited biochemical and molecular information as well.

In their work, published in Nature Cell Biology, the researchers conducted measurements using a refractive index—an intrinsic quantity governing light/matter interaction. Specifically, they encoded 3D refractive index tomograms. Next, that information was decoded with a deep learning (AI)-based model that implies multiple 3D fluorescence tomograms from the refractive index measurements of the corresponding subcellular targets, which achieves multiplexed micro-tomography.

Constant quantitative relationship

The quantitative relationship between the spatial distribution of the refractive index, discovered via AI and the major structures in cells, helped decipher the spatial distribution of the refractive index. The process avoids making subsequent changes to the cell structures, photobleaching, and phototoxicity, while still observing various types of cellular structures in their natural state in 3D and at the same time as fast as one millisecond. The researchers say this also allows long-term measurements over several days.

Fluorescence images can be “directly and precisely predicted from holotomographic images in various cells and conditions,” according to the researchers. The team’s new concept microscope “combines the advantages of several microscopes with the multidisciplinary research of AI, optics, and biology,” says YongKeun Park, professor at KAIST.

The researchers note that “full 3D modeling of absolute and unbiased [refractive index] improves generalization, such that the approach is applicable to a broad range of new samples without retraining to facilitate immediate applicability.”