NCI Uses AI to Take the Guesswork Out of Assessing Prostate Cancer Images
If you’re diagnosing prostate cancer and trying to predict its aggressiveness, you know how vital it is to accurately assess tumor margins—that area between normal healthy tissue and cancer cells. But assessing those margins is an inexact science, often left to the “eye of the beholder.”
Using an artificial intelligence (AI)-based approach, NCI researchers looked at a key magnetic resonance imaging (MRI) component—the T2 weighted image, or “T2WI”—that could help you in better assessing prostate cancer.
Read the full report, “Deep Learning-Based Quality Assessment: Impact on Detection Accuracy of Prostate Cancer Extraprostatic Extension on MRI,” in Abdominal Radiology. You can access the code on GitHub.
The researchers compared images to a pathologist’s assessment of extraprostatic extension (EPE)—a measurement that assesses the number of cells that have spread beyond the prostate borders. The researchers attained a near-perfect accuracy rating in the testing phase, with 85% accuracy in the development stage, making the model an ideal candidate for further clinical evaluation.
In short, their study shows that using AI to assess T2WI quality gives you both an objective and automated way to examine EPE.
According to corresponding author, Dr. Baris Turkbey, of NCI’s Center for Cancer Research’s Molecular Imaging Branch, “Our AI model offers another tool to help predict if cancer will spread or re-occur, giving us vital information for monitoring disease and tailoring treatment to each patient.”
“In the future, a model such as this also could give technicians real-time feedback on the quality of their scans, allowing them to immediately re-scan if needed, saving time and additional resources,” he added.