Bidirectional human-AI collaboration in brain tumour assessment improves both expert human and AI agent performance
Year: 2026
Type: article
Abstract: Artificial intelligence (AI) is increasingly evaluated for its capacity to enhance expert human performance, yet the inverse paradigm, an AI agent supported by a human expert, is rarely studied.Here we investigate both directions of human-AI collaboration in brain tumour imaging assessment.Eleven board-certified radiologists assess 564 cases across ten datasets and four countries in a prospective, multi-reader randomised crossover reader study, predicting contrast enhancement from pre-contrast magnetic resonance imaging alone, with and without AI segmentation support.Accuracy and confidence, and their alignment, improve for both agents when each supports the other.AI support raises radiologist balanced accuracy from 0.698 to 0.743 and reduces reporting time from 45.6 to 30.6 seconds per case.Radiologist input raises the AI's balanced accuracy from 0.824 to 0.842 and its confidence-accuracy correlation from -0.078 to 0.272.Synergistic gains arise when human and AI support one another, underscoring the value of bidirectional augmentation over replacement.
Institutions: National Hospital for Neurology and Neurosurgery, UCL Queen Square Institute of Neurology, University College London, Nvidia (United Kingdom), Great Ormond Street Hospital, Royal National Orthopaedic Hospital, The Royal Free Hospital, Imperial College Healthcare NHS Trust, Imperial College London, University College Hospital, University College London Hospitals NHS Foundation Trust
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7226045663