We've updated our Privacy Policy to make it clearer how we use your personal data.

We use cookies to provide you with a better experience. You can read our Cookie Policy here.

Advertisement
Rectangle Image
Industry Insight

AI in Microscopy: Opportunities, Challenges and the Future

Rectangle Image
Industry Insight

AI in Microscopy: Opportunities, Challenges and the Future

Biological image processing and analysis can often be laborious and complex tasks for researchers. Aivia aims to help researchers tackle the most challenging imaging applications using artificial intelligence (AI)-guided image analysis and visualization solutions.

Technology Networks
spoke with Dr. Luciano Lucas, director at Leica Aivia, to learn more about the challenges of biological image analysis and how AI can help to overcome them. In this interview, Dr. Lucas also explains some of the potential barriers to wider adoption of AI in laboratories and shares his views on where AI microscopy may be headed in the future.

Anna MacDonald (AM): What challenges do researchers face when undertaking biological image analysis? How can AI help to address these issues?

Dr. Luciano Lucas (LL):
Researchers in the biopharma/life sciences spaces are faced with a wide range of problems when it comes to image processing and image analysis. The key issues we have identified and are working on) are:

1) Development, implementation and accessibility to state-of-the-art AI technology (AI microscopy). This type of technology enables the completion of previously impossible to run experiments. However, AI microscopy is a new discipline which requires further research, validation and characterization. We and others in the community are very active on this important task. After four and a half years of R&D we feel confident enough to release to the public some of the work we have done as either pre-trained deep learning models (see our 3D RCAN paper and Aivia DL Model Library) or enabling software tools that allows everyone to leverage some key AI microscopy technology (e.g. AiviaCloud).

2) Inherent image quality. Image acquisition and image analysis are decoupled from each other in time. This often results in the creation of large amounts of image data that are not “good enough” for analysis.

3) Data size. This poses all sorts of issues on both the visualization and analysis fronts.

4) Result accuracy and reproducibility. Both are an essential part of the scientific discovery process.

5) Tool complexity. Making tools that are easy to learn and use is essential to adoption – this is often under appreciated.

Our research work is primarily focused on item one above, but we also have active internal R&D projects to address the rest. As we address the key topics mentioned above we strive to improve the rate of scientific discovery based on image data. We believe this can be achieved by improving how we (humans) interact with software and hardware. Present day tools ignore the fact that researchers are experts in biology (or similar disciplines) and may have very limited expertise in microscopy, image analysis and/or data science/machine learning (ML)/deep learning (DL)/AI. By creating tools that acknowledge and leverage the biologist’s expertise we can create intelligent tools that learn (about biology) from the user. Such tools would gradually learn what a cell is and what it can look like in multiple scenarios. Ultimately, the software/hardware should be able to autonomously do the imaging and image analysis, thus allowing the researcher to focus on the creative and critical thinking portion of the scientific discovery process.

AM: How easy is it for laboratories to adopt AI? Are there any barriers that need to be overcome?

LL:
From the point of view of availability, it is easy. Aivia is a key example of a professionally developed and supported software platform that can be used by anyone. There are several open source projects that offer powerful technical solutions in this space too. The issue/problem for wider adoption is tool and technology complexity. AI is a new topic within the microscopy/biomedical sciences community. Thus, there are very few experts and fewer good tools.

In the last three years we have seen a major increase in pre-prints and peer-reviewed publications using AI for microscopy as well as the creation of several high-profile courses and symposia on the topic (see the AI Microscopy Symposium). I expect the number of publications to continue to increase in the coming years as this type of approach splits out of the labs/groups/companies that have been pioneering it and become “mainstream”. The leaders in this community will need to continue their outreach and educational activities – in turn this will help solve the key issues mentioned above.

It is key to create tools (software and hardware) that clearly show the value of AI for microscopy. Today’s best AI-powered tools can achieve a lot in the hands of ML/DL experts but, for the most part, are not easy to use for non-experts in this space. Our team is very aware of this and is focused on creating tools (Aivia/AiviaWeb/AiviaCloud) that remove the complexity while delivering the full power of AI for microscopy applications.  

AM: Can you tell us more about Aivia and what sets it apart?

LL:
Aivia makes AI microscopy accessible to all. From image restoration (and super resolution) to image segmentation and virtual staining, we can do it all in one easy-to-use platform. Aivia is also great for large (multi TB) data sets and has several good solutions for automation and reproducibility.

AM: What do you see in store for the future of AI in microscopy?

LL:
It is a true pleasure to work in this field – nearly every day one comes across new ideas with significant potential to be transformative. Below are some of my favorites (not all of them used in the microscopy world – at least not yet).

·         GPT3

·         Transformers

·         Flood filling networks

·         Smart microscopy

·         U-net

·         CARE

·         Neuromorphic processing units

·         Optical deep learning

·         Virtual staining


In the next few decades, we will gradually move from AI solutions/tools that are good (i.e. human level performance) at “System 1” learning and thinking, to AI agents that can do “System 2” learning and thinking. This is the true challenge both in microscopy and more generally. Humans will likely remain far superior to AI agents for tasks that require the integration and consideration of multiple, incomplete, multi-domain data sources. As we create better AI agents that can act more often in a System 2 way, humans will be able to dedicate more of their time to creative and innovative tasks, e.g. creating scientific hypotheses, designing experiments to test those and interpreting the insights provided by the AI agents.

Dr. Luciano Lucas was speaking to Anna MacDonald, science writer for Technology Networks.

Meet The Author
Anna MacDonald
Anna MacDonald
Science Writer
Advertisement