Rapid Assisted Visual Search: Supporting Digital Pathologists with Imperfect AI
summary
An interactive review tool pairs a convolutional neural network that scores cancer probability per pixel (section score = max pixel) with an automated RSVP-style navigator and fixed four regions per section chosen to achieve 100% sensitive detection on the test set. A within-subject study with six pathologists shows time drops from ~17 to ~6 minutes per case with equal or slightly higher sensitivity, but users struggle to detect model malfunctions and trust grows only after repeated negative confirmations.
AI-written summary · about
related talks
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
24mBreast cancer detection with IBM PowerAI Vision
21mFighting Cancer with AI
25mComputer Vision in Melanoma Detection
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
21mCreating Markets, Creating Trustworthiness: Standards for AI
29mWhat do a tree and the human brain have in common-a not so serious introduction to digital pathology
55mDeep Learning in Medical Image Diagnostics