Zhi Huang1,2, Eric Yang1, Jeanne Shen1, Dita Gratzinger1, Frederick Eyerer1, Brooke Liang1, Jeffrey Nirschl1, David Bingham1, Alex M. Dussaq1, Christian Kunder1, Rebecca Rojansky1, Aubre Gilbert1, Alexandra L. Chang-Graham1, Brooke E. Howitt1, Ying Liu1, Emily E. Ryan1, Troy B. Tenney1, Xiaoming Zhang1, Ann Folkins1, Edward J. Fox1, Kathleen S. Montine1, Thomas J. Montine1,*, James Zou2,*
1 Department of Pathology, Stanford University School of Medicine, Stanford, CA 94305, USA
2 Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA
Nuclei.io enables pathologists to rapidly create diverse datasets and models for multiple clinically critical applications. To validate the effectiveness of this framework, we perform two cross-over user studies using a pathologist–AI collaborative approach. In both studies, the use of AI yielded considerable improvements in sensitivity, certainty, and efficiency in pathological diagnosis. Our findings validate the proposed human-in-the-loop framework, and suggest that pathologist–AI collaboration holds the potential to revolutionize digital pathology, ultimately leading to improved quality, efficiency, and cost.
The dataset (with RGB image patch and label) that annotated by pathologists.