LHO

The Ledger · Entry 096 · Perception

Finding a metastasis in a lymph node slide

ContestedSince 2017Criterion performanceLast reviewed 2026-08-17

Event

In December 2017 JAMA published the results of the CAMELYON16 challenge, in which several algorithms read 129 whole slide images more accurately than a panel of 11 pathologists working under the time limits of routine practice.

The task is narrow, finding deposits of breast cancer in sentinel lymph node sections, and the comparison was a simulation built to imitate a normal workload. The best algorithm was described by the authors as comparable to a single pathologist given unlimited time, not better than one. A metastasis found on a slide is a fact, not a report: staging the disease, taking the case to the tumour board and choosing what to give the patient afterwards were all outside the exercise. What was measured is the detection of a small object on a large image. What was not measured is the signature under the diagnosis.