AnnotationBridgeBack to site

Medical image annotation for research labs

AnnotationBridge prepares student volunteers for defined image annotation tasks. Research teams provide the task instructions and review submitted work.

Tasks and review

Volunteers can trace visible structures and label specified features in de-identified medical images.

Volunteers do not make clinical decisions. Researchers approve the annotations, and students receive volunteer hours for accepted work.

Relevant research

These studies examine how non-experts perform defined image annotation tasks with training or expert comparison.

Dermoscopic skin lesion examples

Dermoscopic skin lesion

Untrained contributors achieved good-to-excellent agreement with experts when identifying six visible dermoscopic features (κ = 0.71–0.84).

Kentley et al., 2023

Fundus photograph of a human retina

Fundus photograph

Ophthalmology-naive annotators produced abnormality labels comparable to experts; combined classifications reached an AUC of 0.93.

Mitry et al., 2016

Chest CT examples with highlighted nodules

Chest CT

After a brief tutorial, non-experts detected lung nodules with over 90% sensitivity, and 96% for nodules larger than 4 mm.

Boorboor et al., 2018

Project process

Volunteer preparation

  1. Training video followed by comprehension questions
  2. Tasks assigned by difficulty based on volunteer experience
  3. Multiple volunteers independently annotate each image
  4. Accuracy measured using Dice, IoU, boundary error, and Hausdorff distance

Research team oversight

  1. Raw imaging data
  2. De-identification
  3. De-identified images only
  4. IRB protocol
  5. Visual tasks only, with no medical decisions
  6. Expert review and final approval authority
  7. Volunteer hours awarded for accepted work only