Impact of test set composition on AI performance in pediatric wrist fracture detection in X-rays
Question Do different sampling strategies based on samples' complexity have an influence in deep learning models' performance in fracture detection? Findings AI performance in pediatric wrist fracture detection significantly drops when tested on balanced datasets with more challenging cases, compared to randomly selected cases. Clinical relevance Without standardized and validated test datasets for AI that reflect clinical complexities, performance metrics may be overestimated, limiting the utility of AI in real-world settings.
Authors
- Mario Scherkl
- Christina Flucher (ORCID: https://orcid.org/0000-0002-9754-3794)
- Georg Singer (ORCID: https://orcid.org/0000-0003-4770-3144)
- Nikolaus Stranger (ORCID: https://orcid.org/0009-0001-7274-9403)
- Franko Hržić (ORCID: https://orcid.org/0000-0003-1513-0337)
- Sebastian Tschauner (ORCID: https://orcid.org/0000-0002-7873-9839)
- Ivan Štajduhar (ORCID: https://orcid.org/0000-0003-4758-7972)
- Tristan Till
- Saskia Hankel
Institutions
- University of Rijeka (HR)
- Medical University of Graz (AT)
Publication Details
- Journal
- European Radiology
- Published
- 2025-05-16
- DOI
- https://doi.org/10.1007/s00330-025-11669-z
- Citations
- 5
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 3.27