Member-only story
Histopathology & Foundation Models: How good are they really?
A Comprehensive Comparison for Classification of Ovarian Cancer
Ovarian cancer has long been one of the most elusive and dangerous forms of cancer affecting women worldwide. Early detection is critical, yet pathologists often face difficulty in distinguishing between subtypes that can look remarkably similar under the microscope. Now, a dedicated team of researchers — Jack Breen, Katie Allen, Kieran Zucker, Lucy Godson, Nicolas M. Orsi, and Nishant Ravikumar — from the University of Leeds and Leeds Teaching Hospitals NHS Trust, among others, have conducted a large-scale study that may transform how pathologists diagnose ovarian cancer. Published in npj Precision Oncology, their work demonstrates that “foundation models” — originally developed using advanced machine learning strategies known as self-supervised learning — could substantially improve ovarian cancer subtype classification.
Why Ovarian Cancer Subtyping?
Understanding the subtype of ovarian cancer is often a matter of life and death. Ovarian tumors show a broad range of morphological (i.e., microscopic) characteristics. Five key subtypes — high-grade serous (HGSC), endometrioid (EC), clear cell (CCC), low-grade serous (LGSC), and mucinous (MC) — are collectively…
