Engineers leverage AI to improve cancer detection, diagnosis
Artificial intelligence, or AI, is impacting our lives every day, from the social media apps on our phones to the smart assistants in our homes. Now, Ohio State engineers are deploying AI in the fight against cancer by improving digital pathology diagnosis.
Digital pathology uses ultra-fast, high-capacity scanners to convert traditional glass slides into high-resolution digital images. These enhanced images can be viewed and interpreted by pathologists on a computer —by multiple people and across multiple locations simultaneously when needed. The images also can be paired with other clinical information to provide a wholly unified picture of each person’s unique cancer.
While digital pathology has been in use for some time, it has grown in prevalence over the past decade, with Ohio State among the leading innovators and users of the technology. With the help of AI and high-performance computing, researchers led by Computer Science and Engineering Professor Raghu Machiraju are taking the technology to new levels.
The engineers are applying AI to improve digital image analysis, complementing the work of the pathologist to achieve greater levels of accuracy and consistency.
“Can we build a model that can completely emulate a pathologist? No way, but can we build a synergy between human and machine to make it easier for the pathologist? That’s what we’re trying to do,” said Machiraju, who also holds appointments in the Departments of Biomedical Informatics and Pathology.
Using data from colleagues at the James Cancer Hospital and Solove Research Institute and other partner institutions, engineers create machine learning models and algorithms that can identify predictive patterns in these images, or that can provide objective and reproducible interpretations of their contents. The machine searching for and identifying cancer criteria or cues can help expedite and inform the work of pathologists.
One way Ohio State engineers are aiding the work of pathologists is by developing an app that measures tumor budding, one of the microscopic features seen in different cancers that may indicate a more aggressive spread. Improving how cancers are classified helps the clinical doctors counsel their patients in a more specific manner, thus helping improve diagnosis, treatment and prognosis.
Successful machine learning is dependent upon large and broad data sets, said Machiraju, which can be challenging to collect. Ohio State is hoping to lead a consortium of partner institutions in the future to address this issue. The goal is to create a large repository which will allow the necessary training of machine learning diagnostic models for various cancers. Initially the team will focus on collecting patient images and data pertaining to breast, prostate, and thyroid cancers, glioblastomas, and sarcomas.
Additional funding support could help provide user training and data collection, both of which are essential to the success of the technology, said Machiraju.
“If we don’t have data to train the machines, AI is pretty useless—it's moot.”
Machiraju said another goal is to make AI more accessible for clinicians across the globe by providing access to cloud resources to those who lack adequate on-premises computing capabilities and skillsets. Ohio State is the home base of two new federally funded institutes dedicated to advancing artificial intelligence research. One of those institutes, the AI Institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE), will build the next generation of cyberinfrastructure with a goal of making AI data and infrastructure more accessible to the larger society. Machiraju is co-Principal Investigator of ICICLE.
by Meggie Biss, College of Engineering Communications | biss.11@osu.edu
