Digital pathology is a new field that primarily deals with microscopy images derived from patient biopsies. Because of the high resolution, most of these whole slide images (WSI) are quite large, often exceeding a gigabyte in size (Gb). As a result, standard image analysis methods cannot handle them efficiently.
A new artificial intelligence (AI) algorithm developed by researchers in Boston was able to classify lung cancer subtypes based on tissue images from resected tumors. The findings highlight the value of digital pathology, a new field that focuses on microscopy images derived from patient biopsies.
Recognizing a need, Boston University School of Medicine (BUSM) researchers created a novel artificial intelligence (AI) algorithm based on a framework called representation learning to classify lung cancer subtypes based on images of resected tumors.
Pathology practice is undergoing a digital revolution. Computer-based methods are being developed to assist the expert pathologist. Furthermore, in places where there is no expert, such methods and technologies can directly assist in diagnosis.
Vijaya B. Kolachalama
“We are working on developing novel AI-based methods for analyzing digital pathology data. Pathology practice is undergoing a digital revolution. To assist the expert pathologist, computer-based methods are being developed. Furthermore, in areas where there is no expert, such methods and technologies can directly assist in diagnosis “Vijaya B. Kolachalama, Ph.D., FAHA, assistant professor of medicine and computer science at BUSM, explains the corresponding author.
“Pathology practice is undergoing a digital revolution,” says corresponding author Vijaya B. Kolachalama, PhD, FAHA, assistant professor of medicine and computer science at Boston University School of Medicine (BUSM). “Computer-based methods are being developed to assist the expert pathologist.” Furthermore, in places where there is no expert, such methods and technologies can directly assist in diagnosis.”
The researchers created the Graph Transformer (GTP), a graph-based vision transformer for digital pathology that uses a graph representation of pathology images and the computational efficiency of transformer architectures to perform analysis on the entire slide image.
“Translating the latest advances in computer science to digital pathology is not straightforward and there is a need to build AI methods that can exclusively tackle the problems in digital pathology,” explains co-corresponding author Jennifer Beane, PhD, associate professor of medicine at BUSM.
They then developed a model that could distinguish between lung adenocarcinoma, lung squamous cell carcinoma, and adjacent non-cancerous tissue using whole slide images and clinical data from three publicly available national cohorts. They demonstrated, through a series of studies and sensitivity analyses, that their GTP framework outperforms current state-of-the-art methods for whole slide image classification.
Their machine learning framework, they believe, has implications beyond digital pathology. “Researchers interested in developing computer vision approaches for other real-world applications may find our approach useful as well,” they added.
The BUSM team then developed a model that could differentiate between lung adenocarcinoma, lung squamous cell carcinoma, and adjacent non-cancerous tissue using whole slide images and clinical data from three publicly available national cohorts. They demonstrated that their GTP framework outperforms current state-of-the-art methods for whole slide image classification in a series of studies and sensitivity analyses. These findings have been published online in the journal IEEE Transactions on Medical Imaging.
