A groundbreaking “deep learning” artificial intelligence (AI) model developed at Washington State University (WSU) is showing promising results in identifying signs of disease in both animal and human tissue. This model, which is faster and often more accurate than human pathologists, could significantly accelerate disease research and improve medical diagnoses, particularly in the early detection of cancers.
Published in Scientific Reports, the study highlights the AI’s ability to analyze pathology images with remarkable speed and precision. For example, the model can detect cancer from biopsy images in just a few minutes—far outpacing the hours of work typically required by human pathologists. According to Michael Skinner, a biologist at WSU and co-author of the study, this AI system has the potential to “revolutionize” medical diagnostics and pathology, providing a crucial tool for both animal and human health analysis.
The AI model was developed by computer scientists Colin Greeley and his advisor, Professor Lawrence Holder, with input from Skinner’s lab, which focuses on epigenetic research. The researchers trained the model using a set of images from Skinner’s studies on molecular signs of disease in tissues such as kidney, testes, ovary, and prostate from rats and mice. The AI was then tested using additional images, including those identifying human breast cancer and lymph node metastasis.
The results were impressive. Not only did the AI model identify disease markers in tissue samples more quickly than previous models, but it also outperformed human pathologists in several instances, uncovering issues that were missed by the trained experts. Holder, a co-author on the paper, explained that the deep learning model has the potential to “identify disease and tissue faster and more accurately than humans.”
Traditional pathology involves meticulous analysis by specially trained personnel, who examine and annotate tissue slides under a microscope, often cross-checking their work to minimize human error. This process can be slow and labor-intensive, sometimes taking a year or more for large studies. Skinner’s epigenetic research, for instance, could take more than a year to analyze molecular changes in tissue samples. However, with the AI model in place, the same research can now be completed in a matter of weeks, significantly accelerating the pace of scientific discovery.
Deep learning, a form of AI that mimics the human brain, goes beyond traditional machine learning methods. It utilizes a network of artificial neurons and synapses to process and analyze data. When the model makes a mistake, it learns from the error through a process called backpropagation, which adjusts the network to avoid repeating the same mistake.
The WSU team designed their deep learning model to handle extremely high-resolution images, often in the gigapixel range, containing billions of pixels. To manage these massive image files, the AI analyzes smaller “tiles” of the images and places them within the context of the larger tissue section at a lower resolution. This process functions similarly to zooming in and out with a microscope, allowing the model to work efficiently while maintaining accuracy.
The AI model is already attracting interest from other researchers. Holder’s team is collaborating with WSU’s veterinary medicine department to apply the technology to diagnose diseases in deer and elk tissue samples. The model’s ability to quickly and accurately identify diseases in animals holds great potential for wildlife research and veterinary care.
In human medicine, the model could have a profound impact on cancer detection and gene-related diseases. As long as annotated images, such as those identifying cancerous cells in tissue, are available for training, the AI can be trained to perform similar diagnostic tasks. This could lead to faster, more accurate diagnoses for a variety of conditions, particularly cancers, where early detection is crucial for successful treatment outcomes.
Holder emphasized the state-of-the-art nature of the deep learning network, which outperformed other systems in comparison tests using various data sets. “The network we’ve designed is state-of-the-art,” Holder said. “We did comparisons to several other systems and other data sets for this paper, and it beat them all.”
The research, which received support from the John Templeton Foundation, was conducted with the assistance of Eric Nilsson, a WSU research assistant professor. Together, the team is advancing the frontier of AI in medicine, setting the stage for a new era of disease detection and research.
The deep learning model developed at WSU represents a major step forward in the application of AI to pathology and medical research. By providing faster, more accurate disease detection, this technology has the potential to revolutionize both veterinary and human healthcare. With its ability to handle high-resolution tissue images and learn from past data, the AI system could become an indispensable tool for scientists and medical professionals alike, speeding up diagnoses and research processes that could otherwise take years.
By Impact Lab
