AI for lymph-node metastasis detection: how should workflow evidence be read?

Phase II CONFIDENT-B: AI-assistance in detection of breast cancer metastases in sentinel lymph nodes

VJOncology · 2024-02-07 · Source

In this English-language SABCS 2023 interview published by VJOncology on YouTube on 7 February 2024, Carmen van Dooijeweert discusses the CONFIDENT-B workflow. The interview predates the full June 2024 paper; it is not personal treatment advice or a regulatory authorization document.

How does AI assist the pathologist?

A whole-slide image is a digitized view of a microscope slide. The studied systems draw the pathologist’s attention to possible metastatic areas in lymph nodes. Three elements need separate assessment: the algorithm’s standalone performance, the performance of the pathologist–AI combination and the laboratory’s confirmation workflow. These are not interchangeable.

What did CONFIDENT-B test?

Published on 27 June 2024, CONFIDENT-B was a prospective, single-center, non-randomized study of 190 breast-cancer sentinel lymph-node specimens. It evaluated how AI assistance affected immunohistochemistry (IHC) use and review time. IHC remained in place for morphologically negative cases. The findings therefore do not mean that IHC was abandoned entirely or that autonomous diagnosis was tested.

What caution does the gastric-cancer study add?

The 2022 Nature Communications study addressed lymph-node metastases in gastric cancer. Six pathologists reviewed 80 slides with and without AI assistance; investigators assessed time, accuracy and counting consistency. The study includes examples in which false AI highlights also led pathologists toward incorrect interpretations. Trusting an annotation map and verifying the tissue finding are therefore distinct steps.

Are digital-system validation and AI validation the same?

The CAP whole-slide imaging guideline addresses validation of diagnostic accuracy and concordance with light microscopy for the intended use of digital systems. This is guidance on the imaging infrastructure; it does not establish that a particular AI metastasis algorithm is validated or has regulatory authorization. A research publication, permission for local clinical use and an institutional implementation decision require separate appraisal.

How should an implementation claim be assessed?

As an editorial appraisal, ask about cancer type, slide preparation, scanner, software version and reference standard together. Changes in review time, false alerts, missed foci and additional staining needs should be monitored separately. Faster review alone does not establish a survival benefit. This framework offers no personal diagnostic or treatment advice and makes no efficacy claim for RAI products.