Virtual tissue staining by deep learning

Projektinformationen

IFLAI

ID Finanzhilfevereinbarung: 101113141

DOI

10.3030/101113141

Projekt abgeschlossen

EK-Unterschriftsdatum 24 April 2023

Startdatum 1 Mai 2023

Enddatum 31 Oktober 2024

Finanziert unter

• European Research Council (ERC)

Gesamtkosten

Keine Daten

EU-Beitrag

€ 150 000,00

Koordiniert durch

GOETEBORGS UNIVERSITET
Sweden

Periodic Reporting for period 1 - IFLAI (Virtual tissue staining by deep learning)

Berichtszeitraum: 2023-05-01 bis 2024-10-31

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

Tissue staining is an essential tool in biomedical and pharmaceutical research, used to highlight specific cell structures for observation and analysis. Traditional chemical staining techniques, however, are often time-consuming, costly, and invasive. These methods can damage or kill cells, making them unsuitable for live-cell imaging or longitudinal studies. As science advances, there is a growing need for more efficient, non-invasive, and scalable approaches to tissue staining.

The IFLAI project set out to address this need by developing a novel deep-learning-powered virtual-staining solution. The objective was to replace conventional chemical staining with a system capable of generating virtually stained images directly from brightfield microscope data. This technology combines a simple optical device with powerful deep-learning software, enabling users to visualize multiple cell components in real-time, without the need for specialized expertise or hazardous chemicals. The expected impact of this innovation extends beyond improved research productivity to fostering safer, more sustainable practices in laboratories and clinics. It also opens new possibilities for studying live cells over time, advancing drug discovery, diagnostics, and personalized medicine.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

Throughout the project, significant advancements were made in both the technological development and the practical application of virtual staining. A deep-learning framework was developed to translate brightfield images into virtually stained ones, bypassing the need for paired datasets of chemically stained and unstained samples. This innovative approach drastically reduces the complexity and cost of training the model, making it highly adaptable to diverse applications in biomedical research and diagnostics.

In response to strong interest from potential clients and investors, we focused on refining the software component of the solution. By prioritizing the development of a robust and versatile deep-learning-powered software, we were able to meet the demand for a scalable, flexible product. The software was further expanded to include additional functionalities, such as cell tracking and segmentation, addressing broader needs in biomedical research and enhancing its appeal to a wider range of users.

An important milestone achieved during the project was the establishment of the startup IFLAI AB (www.iflai.com) which will spearhead the commercialization of the technology. The startup has enabled the project team to engage directly with stakeholders, including research institutions and industry partners, to validate and refine the solution. These interactions not only informed the technical development but also strengthened the foundation for the product’s market entry.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The IFLAI project has delivered advancements that go beyond the current state of the art in virtual staining and biomedical imaging. Unlike existing methods, which often rely on complex optical setups or require extensive paired datasets, the IFLAI solution simplifies both the hardware and software components. The innovative use of CycleGAN-based deep-learning models eliminates the need for paired data, making it easier and more cost-effective to adapt the technology to new staining applications and diverse imaging setups.

This approach significantly lowers the barriers to adopting virtual staining, offering a flexible and scalable solution for end-users. Additionally, the software's functionality has been expanded to include cell tracking and segmentation, addressing a broader range of challenges in biomedical research. These enhancements make IFLAI not only a tool for virtual staining but also a versatile platform for applications across diagnostics, drug discovery, and personalized medicine.

The project has also laid the groundwork for commercial success through the establishment of the startup IFLAI AB. The company is actively pursuing partnerships, market validation, and intellectual property protection to ensure a smooth pathway to adoption. To further accelerate uptake, efforts are focused on engaging with industry stakeholders, aligning with regulatory standards, and exploring new markets where the technology can create significant value.

Example of virtual staining of a tissue.