In the realm of life sciences, confocal fluorescence microscopy (CFM) is well-known for its ability to create high-resolution images of cells. However, this technique relies on fluorescent staining, which can lead to issues like photobleaching and phototoxicity that may harm the cells being examined. On the other hand, mid-infrared photoacoustic microscopy (MIR-PAM) allows for imaging without labels, helping to maintain the integrity of the cells. Unfortunately, its use of longer wavelengths restricts spatial resolution, making it challenging to accurately visualize detailed cellular features.
Within the field of life sciences, confocal fluorescence microscopy (CFM) stands out for offering high-resolution images of cellular structures. Nonetheless, it necessitates fluorescent dyes which can introduce risks such as photobleaching and phototoxicity, potentially jeopardizing the health of the cells. In contrast, mid-infrared photoacoustic microscopy (MIR-PAM) enables imaging without the need for labels, thus preserving cell integrity. However, its dependence on longer wavelengths results in lower spatial resolution, complicating the precise visualization of delicate cellular components.
To address these limitations, a research group from POSTECH has created a pioneering imaging technique using explainable deep learning (XDL). This method enhances low-resolution, label-free MIR-PAM images into high-resolution images that mimic the quality seen in CFM. Unlike traditional AI approaches, XDL improves transparency by allowing observation of the transformation procedure, which increases the trustworthiness and accuracy of results.
The team utilized a single-wavelength MIR-PAM system and established a two-step imaging process: (1) The Resolution Enhancement phase upgrades low-resolution MIR-PAM images to high-resolution ones, allowing clear visibility of intricate cell structures such as nuclei and filamentous actin; (2) The Virtual Staining phase generates images that appear stained without using fluorescent tags, thus avoiding the hazards linked with staining while achieving CFM-quality imaging. This groundbreaking technology enables high-resolution, virtually stained cellular imaging, safeguarding cell viability and advancing tools for live-cell analysis and detailed biological studies.
Professor Chulhong Kim stated: “We have engineered a transformative technology that connects the physical boundaries of various imaging techniques and offers complementary advantages. The XDL method has greatly improved the stability and dependability of unsupervised learning.” Professor Jinah Jang highlighted, “This study opens up fresh avenues for multiplexed, high-resolution cellular imaging without the need for labels. It possesses significant potential for applications in live-cell studies and research on disease models.”
This research was supported by several organizations, including the Ministry of Education, the Ministry of Science and ICT, the Korea Medical Device Development Fund, the Korean Fund for Regenerative Medicine, the Korea Institute for Advancement of Technology (KIAT), the Artificial Intelligence Graduate School Program at POSTECH, BK21 FOUR, and the Glocal University 30 Project.
