A “deep learning” artificial intelligence model developed at Washington State University is capable of recognizing pathology, or disease indicators, in images of both animal and human tissue more quickly and often more accurately than humans. This advancement has the potential to significantly accelerate disease-related research and improve medical diagnostics, such as detecting cancer in biopsy images within minutes, a task that usually takes several hours for human pathologists.
A “deep learning” artificial intelligence model developed at Washington State University can identify pathology, or signs of disease, in images of animal and human tissue much faster, and often more accurately, than people.
The details of this development are published in Scientific Reports, and it could greatly enhance the speed of research related to diseases. Additionally, it shows promise for advancing medical diagnoses, like diagnosing cancer from a biopsy within minutes, as opposed to the several hours it normally takes a human pathologist.
Michael Skinner, a biologist at WSU and co-corresponding author of the study, stated, “This AI-based deep learning program was very, very accurate at looking at these tissues. It could revolutionize this type of medicine for both animals and humans, essentially better facilitating these kinds of analysis.”
The AI model was developed by computer scientists Colin Greeley, a former graduate student at WSU, and his advisor, Lawrence Holder. They trained the model using images from previous epigenetic studies conducted by Skinner’s lab, which focused on molecular signs of disease in kidney, testes, ovarian, and prostate tissues in rats and mice. The team also tested the AI with images from additional studies that looked into breast cancer and lymph node metastasis.
The research revealed that this new AI deep learning model not only identified pathologies quickly but did so faster than earlier models and, in some instances, detected issues that a skilled human team might overlook.
Holder remarked, “I think we now have a way to identify disease and tissue that is faster and more accurate than humans.”
Traditionally, analyzing tissues requires meticulous work by specially trained teams who examine and annotate slides under a microscope, often needing to cross-check their assessments to minimize errors.
In Skinner’s epigenetic research—which investigates changes in molecular processes affecting gene behavior without altering the DNA itself—this process could extend over a year or more for extensive studies. However, with the new AI model, they can acquire the same data in just a couple of weeks, according to Skinner.
Deep learning is an AI technique aiming to replicate human brain function, surpassing traditional machine learning methods. Holder explained that a deep learning model consists of a network of neurons and synapses, enabling it to learn from mistakes through a process called backpropagation, which adjusts numerous connections in the network to correct errors and prevent recurrence.
The WSU deep learning model was crafted to manage ultra-high-resolution gigapixel images, which contain billions of pixels. To accommodate the massive file sizes that can burden even advanced computers, the researchers engineered the AI to analyze smaller tiles while still considering their context within larger sections at lower resolutions, akin to zooming in and out with a microscope.
This deep learning model has garnered interest from other researchers, and Holder’s team is presently collaborating with WSU veterinary medicine experts to diagnose diseases in deer and elk tissue samples.
Moreover, the authors highlight the model’s promise for enhancing research and diagnosis in humans, especially concerning cancer and other gene-related illnesses. If adequate data, such as annotated images highlighting cancer in tissues, is available, researchers could train the AI to perform such analyses, according to Holder.
Holder remarked, “The network that we’ve designed is state-of-the-art. We did comparisons to several other systems and other data sets for this paper, and it beat them all.”
This study received backing from the John Templeton Foundation. Eric Nilsson, a WSU research assistant professor in the School of Biological Sciences, also contributed as a co-author on this paper.
