The Pathology Images of Scanners and Mobilephones (PLISM) dataset was created by Ochi et al. in 2024 to evaluate AI model robustness to inter-institutional domain shifts. All histopathological specimens were sourced from patients diagnosed and treated at the University of Tokyo Hospital between 1955 and 2018. PLISM-wsi consists of consecutive slides digitized under 7 different scanners.
Use Cases
- Evaluating AI model robustness to scanner-induced domain shifts based on images digitized under 7 different scanners
- Benchmarking domain adaptation techniques for histopathological image analysis based on inter-institutional data
- Training models to generalize across different medical imaging hardware based on the multi-scanner dataset
- Studying temporal trends in pathology specimens based on data spanning from 1955 to 2018
Strengths
- Created specifically for evaluating AI robustness to domain shifts, a key challenge in medical AI
- Specimens sourced from a single institution (University of Tokyo Hospital) over a long period (1955-2018), providing consistency in source material
- Includes images digitized under 7 different scanners, enabling study of hardware variation effects
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download
- Column-level documentation is absent; field semantics must be inferred after download
- Row count and file formats are unknown, which may limit suitability assessment
Provenance
- Source
- University of Tokyo Hospital
- Collection Method
- Histopathological specimens digitized under multiple scanners
- Time Range
- 1955 to 2018
- Freshness
- Last updated 2025-03-03 13:34:17; freshness should be verified
- Geography
- Tokyo, Japan (inferred from hospital location)