About Critical Values

The artificial intelligence (AI) revolution has arrived with promises to change every industry, specialty, and field, including pathology. AI is already being used in medical laboratories around the country, and while it still has a long way to go to truly change how healthcare is delivered, it has already influenced the laboratory in myriad ways.  

For starters, AI is becoming more easily available at a cheaper price point, because of “the democratization of the software tools and methods,” combined with improved algorithms, according to Ulysses Balis, MD, FASCP, Fellow AIMB, and the James French Professor of Pathology Informatics and Associate Chief Medical Information Officer at the University of Michigan. “Algorithms that would be impossible to run even 10 years ago because it would've been too expensive, now we have the memory and the computers to run these algorithms.” 

Along with this increased capacity comes the potential for transformation in healthcare. 

Here are 7 ways AI changed the laboratory in 2024.

1. Accelerating productivity 

AI is finding its place as a laboratory management productivity tool. “You can’t use it to replace pathologists or lab management, but it can accelerate the pace at which work gets done,” he Dr Balis says. 

Some of the areas where it improves productivity includes helping prepare strategic plans, drafting reports, analyzing whether to send out a test or keep it in house, building the test mix for a reference panel, note taking, and much more.  

“These large language models (LLMs) have enough current medical knowledge to assist with even relevance and justification,” Dr. Balis explains.

2. Improving workflow efficiency 

AI is also helping people improve workflow efficiency in the laboratory as it relates to how specimens are assigned, scheduled, and tested, according to Amrom Obstfeld, MD, PhD, Associate Professor of Clinical Pathology and Laboratory Medicine at the Perelman School of Medicine at the University of Pennsylvania. 

AI allows laboratory professionals to “get a sense for when and where the work is coming from” more easily, to assign resources to customize the workforce to the work that needs to get done, Dr. Obstfeld explains. 

Additionally, for multidisciplinary specialties, like oncology, AI can help integrate a lot of varied information from sources such as hematology, anatomic pathology, flow cytometry, plus molecular and cytogenetic information, he says.

3. Improving pre-analytical processes 

On the clinical pathology side , AI is helping with what Dr. Obstfeld calls, “pre-analytical processes.” For example, using CBC data to identify specimens that might be mislabeled i.r. “wrong blood in tube” or specimens that have been contaminated with line fluids.  

AI is even helping some laboratories figure out how and why these problems are happening so clinicians can streamline the processes. Though, these are not necessarily happening at an industry or even institution-wide level. “These are bespoke things that people are developing on their own, in their own laboratories at this point,” Dr. Obstfeld says.

4. Turning laboratory professionals into data scientists 

Because of the democratization of AI tools like Python, Jupiter Notebooks, TensorFlow, etc, “AI enables a laboratory professional, essentially with a little bit of training, to become a very effective data scientist,” Dr. Balis says. 

While the proper information systems and data science tools haven’t yet reached their full potential, Dr. Balis feels that “the writing is on the wall for the opportunity for precision reference intervals,” he says.  

In fact, pathology will emerge as “one of the major supercomputing-based specialties, Dr. Balis says. “We will have a tremendous need for either direct stewardship of the data ourselves as pathologists or a partnership with data scientists, because a lot of what we do will be numerical and computational.” 

5. Improving diagnosis and even prognosis 

Digital pathology has already made diagnosis more efficient. However, AI is starting to improve prognosis as well, Dr. Balis says. In other words, thanks to a combination of quantitative data science, digital images, molecular data, and laboratory data, AI is creating the capability for “predictive power,” Dr. Balis says. 

“These semi-quantitative predictions that anatomic pathology currently makes can be fine-tuned and honed to an extremely sharp set of statements about the true biologic potential of disease, where we can say with a pretty high degree of confidence what the biologic potential of disease is in a specific patient as opposed to say, well, your general disease tends to this,” he explains. Moreover, “It'll hone our diagnostic precision to a point,” says Dr. Balis.

6. Laying the groundwork for multimodal AI models 

Even more exciting prospects of AI lie down the road. “If we really want to tap into the potentials of AI, we have to go multimodal,” according to Hamid Tizhoosh, PhD, Professor of Biomedical Informatics, in the Department of AI and Informatics at The Mayo Clinic in Minnesota. 

While the AI algorithms are not yet equal to a “super-pathologist” who can diagnose or prognosticate based on numerous types of data, from pathology to laboratory tests and everything in between, according to Dr. Tizhoosh that is the direction the research is moving in. To truly get there will require a more unified way to centralize, store, and protect patient health information that is currently dispersed across multiple institutions.

7. Building the possibility of pathology foundation models  

Another exciting influence is the possibility of pathology foundation models—in other words a “gigantic AI model that can learn everything you just adjust it for the downstream tasks such as triage, diagnosis, survival rate prediction, quality control, cellular segmentation, and so on,” Dr. Tizhoosh says. These kinds of models require enormous data input and training. At present, he says, nobody is there yet, though some institutions are making headway on specialty models for a specific organ or disease, such as breast cancer. 

While there is still a long way to go toward AI working to its fullest capacity, Dr. Balis calls it an “exciting time for AI’s role in pathology.” He adds that pathology is entering “a new renaissance period of enlightenment” where researchers and pathologists alike, are discovering ways to combine multiple classes of data we currently generate and extract derivative types of data that would not be possible without computational tools.