Is a score enough? Pitfalls and solutions for AI severity scores

Abstract Severity scores, which often refer to the likelihood or probability of a pathology, are commonly provided by artificial intelligence (AI) tools in radiology. However, little attention has been given to the use of these AI scores, and there is a lack of transparency into how they are generat...

Full description

Saved in:
Bibliographic Details
Main Authors: Michael H. Bernstein, Marly van Assen, Michael A. Bruno, Elizabeth A. Krupinski, Carlo De Cecco, Grayson L. Baird
Format: Article
Language:English
Published: SpringerOpen 2025-07-01
Series:European Radiology Experimental
Subjects:
Online Access:https://doi.org/10.1186/s41747-025-00603-z
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849238179784163328
author Michael H. Bernstein
Marly van Assen
Michael A. Bruno
Elizabeth A. Krupinski
Carlo De Cecco
Grayson L. Baird
author_facet Michael H. Bernstein
Marly van Assen
Michael A. Bruno
Elizabeth A. Krupinski
Carlo De Cecco
Grayson L. Baird
author_sort Michael H. Bernstein
collection DOAJ
description Abstract Severity scores, which often refer to the likelihood or probability of a pathology, are commonly provided by artificial intelligence (AI) tools in radiology. However, little attention has been given to the use of these AI scores, and there is a lack of transparency into how they are generated. In this comment, we draw on key principles from psychological science and statistics to elucidate six human factors limitations of AI scores that undermine their utility: (1) variability across AI systems; (2) variability within AI systems; (3) variability between radiologists; (4) variability within radiologists; (5) unknown distribution of AI scores; and (6) perceptual challenges. We hypothesize that these limitations can be mitigated by providing the false discovery rate and false omission rate for each score as a threshold. We discuss how this hypothesis could be empirically tested. Key Points The radiologist-AI interaction has not been given sufficient attention. The utility of AI scores is limited by six key human factors limitations. We propose a hypothesis for how to mitigate these limitations by using false discovery rate and false omission rate. Graphical Abstract
format Article
id doaj-art-dea8791f126b4a38b791ef0d87ff481c
institution Kabale University
issn 2509-9280
language English
publishDate 2025-07-01
publisher SpringerOpen
record_format Article
series European Radiology Experimental
spelling doaj-art-dea8791f126b4a38b791ef0d87ff481c2025-08-20T04:01:43ZengSpringerOpenEuropean Radiology Experimental2509-92802025-07-01911510.1186/s41747-025-00603-zIs a score enough? Pitfalls and solutions for AI severity scoresMichael H. Bernstein0Marly van Assen1Michael A. Bruno2Elizabeth A. Krupinski3Carlo De Cecco4Grayson L. Baird5Department of Diagnostic Imaging, Brown Radiology Human Factors Lab, Rhode Island Hospital, Warren Alpert School of Medicine of Brown UniversityDepartment of Radiology and Imaging Sciences, Emory University, School of MedicinePenn State College of Medicine, The Milton S. Hershey Medical Center, Penn State HealthDepartment of Radiology and Imaging Sciences, Emory University, School of MedicineDepartment of Radiology and Imaging Sciences, Emory University, School of MedicineDepartment of Diagnostic Imaging, Brown Radiology Human Factors Lab, Rhode Island Hospital, Warren Alpert School of Medicine of Brown UniversityAbstract Severity scores, which often refer to the likelihood or probability of a pathology, are commonly provided by artificial intelligence (AI) tools in radiology. However, little attention has been given to the use of these AI scores, and there is a lack of transparency into how they are generated. In this comment, we draw on key principles from psychological science and statistics to elucidate six human factors limitations of AI scores that undermine their utility: (1) variability across AI systems; (2) variability within AI systems; (3) variability between radiologists; (4) variability within radiologists; (5) unknown distribution of AI scores; and (6) perceptual challenges. We hypothesize that these limitations can be mitigated by providing the false discovery rate and false omission rate for each score as a threshold. We discuss how this hypothesis could be empirically tested. Key Points The radiologist-AI interaction has not been given sufficient attention. The utility of AI scores is limited by six key human factors limitations. We propose a hypothesis for how to mitigate these limitations by using false discovery rate and false omission rate. Graphical Abstracthttps://doi.org/10.1186/s41747-025-00603-zArtificial intelligenceBiasCognitionRadiologyReproducibility of results
spellingShingle Michael H. Bernstein
Marly van Assen
Michael A. Bruno
Elizabeth A. Krupinski
Carlo De Cecco
Grayson L. Baird
Is a score enough? Pitfalls and solutions for AI severity scores
European Radiology Experimental
Artificial intelligence
Bias
Cognition
Radiology
Reproducibility of results
title Is a score enough? Pitfalls and solutions for AI severity scores
title_full Is a score enough? Pitfalls and solutions for AI severity scores
title_fullStr Is a score enough? Pitfalls and solutions for AI severity scores
title_full_unstemmed Is a score enough? Pitfalls and solutions for AI severity scores
title_short Is a score enough? Pitfalls and solutions for AI severity scores
title_sort is a score enough pitfalls and solutions for ai severity scores
topic Artificial intelligence
Bias
Cognition
Radiology
Reproducibility of results
url https://doi.org/10.1186/s41747-025-00603-z
work_keys_str_mv AT michaelhbernstein isascoreenoughpitfallsandsolutionsforaiseverityscores
AT marlyvanassen isascoreenoughpitfallsandsolutionsforaiseverityscores
AT michaelabruno isascoreenoughpitfallsandsolutionsforaiseverityscores
AT elizabethakrupinski isascoreenoughpitfallsandsolutionsforaiseverityscores
AT carlodececco isascoreenoughpitfallsandsolutionsforaiseverityscores
AT graysonlbaird isascoreenoughpitfallsandsolutionsforaiseverityscores