The framework was able to accurately classify different lung tumors
Digital pathology is an emerging field that primarily deals with microscopy images derived from patient biopsies. Due to the high resolution, most of these whole slide images (WSI) are large in size, typically exceeding one gigabyte (Gb). Therefore, typical image analysis methods cannot handle them effectively.
Seeing a need, researchers at Boston University School of Medicine (BUSM) developed a new artificial intelligence (AI) algorithm based on a framework called representation learning to classify lung cancer subtype based on images of lung tissue from resected tumors.
“We are developing new AI-based methods that can bring efficiency to the evaluation of digital pathology data. The practice of pathology is undergoing a digital revolution. Computational methods are being developed to help the expert pathologist. Moreover, in places where there are no experts, such methods and technologies cannot directly assist in diagnosis,” says corresponding author Vijaya B. Kolachalama, PhD, FAHA, Assistant Professor of Medicine and IT at BUSM.
Researchers have developed a graph-based vision transformer for digital pathology called Graph Transformer (GTP) that leverages a graphical representation of pathology images and the computational efficiency of transformer architectures to perform whole-image analysis of the slide.
“Translating the latest advances in computer science into digital pathology is not straightforward, and there is a need to create AI methods that can tackle digital pathology problems exclusively,” says co-corresponding author Jennifer Beane, PhD, associate professor of medicine at BUSM.
Using whole-slide images and publicly available clinical data from three national cohorts, they then developed a model that could distinguish between lung adenocarcinoma, lung squamous cell carcinoma, and adjacent non-cancerous tissue. In a series of studies and sensitivity analyses, they showed that their GTP framework outperforms current state-of-the-art methods used for whole-slide image classification.
They believe their machine learning framework has implications beyond digital pathology. “Researchers interested in developing computer vision approaches for other real-world applications may also find our approach useful,” they added.
These results are published online in the journal IEEE Transactions on Medical Imaging.
Funding for this study was provided by grants from the National Institutes of Health (R21-CA253498, R01-HL159620), Johnson & Johnson Enterprise Innovation, Inc., American Heart Association (20SFRN35460031), Karen Toffler Charitable Trust and National Science Foundation (1551572, 1838193)