Artificial Intelligence Models Help Detect and Diagnose Cancer

United States

This year, HJF researchers helped develop two groundbreaking artificial intelligence (AI) models for detecting and diagnosing cancer. The models each provide further evidence of the transformative potential impact of AI on health care. “The addition of AI in a physician’s toolkit is similar to the introduction of calculators for engineers,” said James Wren, a Data Scientist on the research team.

One model demonstrates high performance in three areas—spatial resolution, grading resolution and outcome prediction—for detecting prostate cancer. Another model detects with unprecedented accuracy metastatic breast cancer that has spread to the lymph nodes. “It is truly rewarding knowing that our team is working on a device that is expected to improve the accuracy of cancer diagnoses,” said Briana Rivas, a Clinical Research Assistant and member of the slide scanning team.

An augmented reality microscope makes it possible to improve the accuracy and efficiency of cancer diagnosis. The microscope has a computer that uses AI to highlight key areas on a pathology slide, which provides doctors with expert decision support in real time.

The groundbreaking work was the result of collaboration with Naval Medical Center San Diego and Google Health. “With each additional milestone we discover other teams working on similar AI projects, solving similar problems, and as we move forward we continue to form new and exciting partnerships,” said Wren. “Navigating these partnerships allows us to move and grow in exciting ways forming the foundation with which everyone can build upon.”

Developing high-quality AI requires massive amounts of detailed and highly accurate data. Working out of Naval Medical Center San Diego, the HJF research team has spent more than two years collecting and scanning tumor biopsy slides at ultra-high resolution from thousands of cases of prostate and breast cancer. Wren developed sophisticated natural language processing software that searches pathology records to locate cancer cases. After cancer cases have been identified in the archive, Rivas and Joshua Pomorski, Research Technician, develop procedures to locate, prepare and scan the slides. The scanned images are sent to expert pathologists, who grade the cancers and score the severity.

The research team plans to continue to process other types of tissue biopsies so that we can develop more models in the future. “Next we are going to digitize many more tissue types commonly affected by cancer,” said Pomorski. “HJF researchers and machines are working day and night, while you sleep, crunching the numbers and getting closer to scientific advances in diagnosis and disease characterization.”