Shared Resource Spotlight: Center for Bioimaging Informatics

Artificial intelligence (AI) and imaging are rapidly advancing fields that are pushing the boundaries of scientific discovery. These fields are colliding and changing so fast that researchers are strained to the brink trying to stay current. Modern microscopes and imaging platforms can generate thousands of images in a single experiment, capturing cellular and molecular details at a scale that would have been difficult to imagine only a generation ago. However, that abundance creates a new challenge: The more researchers can see, the harder it becomes to analyze what they have captured.

At St. Jude, the Center for Bioimaging Informatics (CBI) helps scientists meet those challenges. The Shared Resource pairs expertise in AI and machine learning to help researchers turn complex image data into biological insight. By collaborating with labs across the institution, CBI makes it possible to answer questions that might otherwise be too time-consuming, technically difficult or computationally demanding to pursue.

Leading the charge is Ali Khalighifar, PhD, CBI director, an expert in using AI to analyze imaging data. For him, the need for that expertise is growing quickly as image acquisition and downstream processing demands outpace traditional image analysis approaches.

“Biologists are now imaging so much — sometimes with thousands of images per experiment — that manual inspection is no longer feasible. Instead, we must use AI and computational approaches,” Khalighifar explained “CBI is sitting at that intersection of microscopy, computation and biology, so image data can be effectively analyzed to answer essential biological questions that were not possible prior to advances in AI.”

Collaborating with CBI for better bioimage analysis

Before providing the best bioimage analysis assistance possible, CBI also needs a proper understanding of the biology involved. At a research institution such as St. Jude, biologists are often some of the top experts in their field, with few peers, if any, who have the same level of knowledge about their chosen topic. Given that CBI has the same level of expertise in AI and machine learning, especially with respect to images, direct collaborations with CBI allow both teams to pool their knowledge to achieve the best outcomes possible.

“We provide as-needed expertise and assistance with standard bioimage analyses, but our most meaningful contributions come through collaboration,” Khalighifar stated. “We are an AI and computational partner, and we want to be involved from the beginning to the end of the research process, informed by our collaborators’ biological expertise, so that we can provide solutions that drive innovation in both biological and machine learning fields.”

The process begins when a scientist contacts CBI with the biological question they wish to answer, the imaging modality they plan to use and the current challenges they are facing. From there, CBI assembles relevant specialists from the team to discuss what support is needed, which can have a variety of potential outcomes.

“Researchers come to us with complex biological questions and imaging data from a wide range of modalities,” said Krishnan Venkataraman, CBI senior image data scientist. “Our role is to transform those images into quantitative, reproducible measurements that reveal biological insight. We build AI-driven analytical tools, models and image processing algorithms to extract meaningful information from imaging data.” 

One such lab that came with a complex imaging task was that of Michael Dyer, PhD, Developmental Neurobiology Department chair. The group needed to image multiple cell types and combine that data, with CBI unlocking their ability to do so. 

“Working with CBI to segment three cell types in order to define interactions and morphology, they have been incredible on all fronts,” Dyer said. “Not only did they innovate new methods to enable detailed morphological analyses, but also ensured these magnificent tools are easy to employ with user-friendly interfaces.”

CBI can help create built-for-purpose image analyses

Using these tools and techniques, CBI can assist with various imaging types and formats. From traditional light microscopy and electron microscopy to in vivo imaging and drug screenings, CBI has helped navigate many imaging challenges at the institution. However, the nature of research is to explore the unknown, so pre-existing tools are often incapable of helping scientists in the ways they need. In these cases, CBI can help create custom analytic packages, which can save large amounts of time and effort.

“In one case, we collaborated with a lab investigating chromocenter locations within a cell’s nucleus in a 3D plane,” said Jaison John, CBI image data scientist. “In the past, they’ve manually looked for specific bright spots in their microscopy, but the number of images involved in a 3D imaging project made that impossible. We automated that process, saving them hours of manual processing per image.”

Another example is helping the lab of Joseph Opferman, PhD, Department of Cell & Molecular Biology. The scientists had discovered a novel effect in mitochondria, but no existing tools could help them analyze their unusual results. “CBI developed a pipeline that allowed us to quantify multiple mitochondrial features, an important aspect that we struggled with before we started working with them,” Opferman said.

For such projects, CBI is most helpful when included early in the project. Consulting at the beginning of a study can help researchers avoid collecting images that are difficult to analyze later or designing experiments that current tools cannot support. With early input, CBI can help labs align image acquisition with the analysis needed to gain true insight.

“With enough information, we can provide guidance from an image analysis perspective before images are acquired,” John explained. “That helps researchers optimize their image acquisition process, so they collect data with the quality and structure needed for downstream analysis. In an ideal interaction, we can also train a lab to apply those principles themselves, so they can operate independently on future analyses using similar conditions.”

CBI trains researchers to analyze images independently

That emphasis on independence is central to CBI’s work. For more common analyses, the team maintains existing tools and documentation and trains researchers to use them. When a consultation reveals that an established tool will meet a lab’s needs, CBI focuses on helping researchers become comfortable enough to carry out the analysis themselves.

While useful, training a single lab at a time like this is too slow to disseminate the latest techniques in imaging analysis across St. Jude, which requires a different approach. To make image-analysis skills more broadly available, CBI provides the Bioimage Analysis Course twice a year to anyone interested. These classes provide the basics of image analysis, especially for non-computational scientists who are beginning to work with increasingly complex imaging datasets.

“We teach researchers how to use established image analysis methods, develop their own computational pipelines and integrate state-of-the-art AI models into their workflows,” explained Venkataraman, who co-instructs the Bioimage Analysis Course with John. “We also teach researchers how to use the tools developed and deployed at CBI, so they can independently apply advanced image analysis methods to their own research questions.”

Maintaining image analysis infrastructure

Through these collaborations and education sessions, the CBI team sometimes notices patterns in what challenges researchers encounter. A recurring problem across multiple labs indicates an unmet need. When that happens, CBI develops a software solution that can be used going forward, expanding the available set of computational tools at the institutional level.

“Many imaging challenges are shared across biological systems and imaging modalities,” Venkataraman said. “When we recognize recurring patterns, we do not simply solve each problem individually. We build reusable computational tools and infrastructure so future researchers can start from a stronger foundation instead of reinventing existing solutions.”

Similarly, CBI keeps track of all previous projects. While still in development, they plan to make the resulting tools available to all researchers, hopefully saving effort by minimizing time spent “reinventing the wheel” if somebody else at St. Jude has already solved the problem.

“We are building a model zoo, which is basically a repository of models that we have built in our previous projects,” John explained. “Researchers can browse this model zoo and then see if anyone has encountered and solved a similar problem; then, they can pull the model and start their project from there instead of starting from scratch.”

CBI looks towards a collaborative future

The pace of change in AI and machine learning can be difficult for any single lab to track, especially when it is not their primary focus. CBI gives St. Jude researchers access to specialists whose work focuses on staying current with those advances and translating them into practical tools for biomedical research. Whether researchers are studying molecules, cells or tissues, they will need computational support to extract meaningful information from images at scale — and CBI is designed to help.

“Almost every lab at St. Jude touches on imaging in some capacity; so, we see our role as collaborators that can enable our colleagues across those many disciplines to analyze images and answer their scientific questions,” Khalighifar concluded. “In the big picture, CBI is a strategic partner that can help researchers navigate the complexities of how best to use emerging tools, including AI, to make the discoveries that will lead to better outcomes for children with catastrophic diseases.”