TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

research

Seventy Percent of the Labels Left. The Scan Model Held

FEEDS selected diverse PET/CT cases for annotation and matched fully labeled training in the authors' multi-dataset evaluation.

Published Updated Story ID: mp-2026-08-12-006
Read the complete editionStory JSON

Summary

FEEDS selected diverse PET/CT cases for annotation and matched fully labeled training in the authors' multi-dataset evaluation.

FEEDS uses foundation-model embeddings to choose informative and diverse unlabeled whole-body scans for expert annotation in one step. Tested across AutoPET-III, DeepPSMA and an internal Dartmouth-Hitchcock dataset, the authors report performance matching training on the full labeled set while reducing annotation burden by 70 percent. The work evaluates segmentation across tracers and cancers; it does not establish clinical deployment or replace expert review.

Why it matters

FEEDS selected diverse PET/CT cases for annotation and matched fully labeled training in the authors' multi-dataset evaluation.

Limits and context

  • The work evaluates segmentation across tracers and cancers; it does not establish clinical deployment or replace expert review.

Key claims

  1. FEEDS selected diverse PET/CT cases for annotation and matched fully labeled training in the authors' multi-dataset evaluation.

    Qualification: The work evaluates segmentation across tracers and cancers; it does not establish clinical deployment or replace expert review.

    Evidence: source-2026-08-12-006

Sources

  1. arXiv preprint 2608.11076arXiv · primary research

Corrections

No corrections have been recorded for this story.