AI in Diagnostic Radiology: What Happens When Algorithms Are Updated

Martine R. Skregelid; Kasim Ibrahim-pur; Flemming Skjøth; Malene Roland Pedersen; Helle Precht

doi:10.3390/radiation6010004

AI in Diagnostic Radiology: What Happens When Algorithms Are Updated

Martine R. Skregelid
, Kasim Ibrahim-pur
, Flemming Skjøth
, Malene Roland Pedersen
, Helle Precht

Radiografuddannelsen Herning

University of Southern Denmark
Vejle Sygehus, Sygehus Lillebælt
University College Cork

Research output: Contribution to journal › Journal article › Research › peer-review

Abstract

Hand and ankle fractures are common and can easily be missed in busy emergency departments. Therefore, many hospitals now use artificial intelligence (AI) to highlight possible fractures. AI software is frequently updated, and regular evaluation is warranted to assesswhether new updates alter the diagnostic performance. In this study, we investigated the diagnostic accuracy of two different versions of the same AI fracture detection software on hand and ankle radiographs, using reporting radiographers’ reports as the reference standard. We also assessed whether AI output led to changes in reporting radiographers’ assessments in cases of disagreement.

Original language	English
Article number	4
Journal	Radiation
Volume	6
Issue number	1
Number of pages	12
ISSN	2673-592X
DOIs	https://doi.org/10.3390/radiation6010004
Publication status	Published - Jan 2026

Keywords

clinical assessment methods, lab technology and radiography

Access to Document

10.3390/radiation6010004Licence: CC BY-ND

Cite this

@article{6488440f543440c08f1db667ac3aa29b,

title = "AI in Diagnostic Radiology: What Happens When Algorithms Are Updated",

abstract = "Hand and ankle fractures are common and can easily be missed in busy emergency departments. Therefore, many hospitals now use artificial intelligence (AI) to highlight possible fractures. AI software is frequently updated, and regular evaluation is warranted to assesswhether new updates alter the diagnostic performance. In this study, we investigated the diagnostic accuracy of two different versions of the same AI fracture detection software on hand and ankle radiographs, using reporting radiographers{\textquoteright} reports as the reference standard. We also assessed whether AI output led to changes in reporting radiographers{\textquoteright} assessments in cases of disagreement.",

keywords = "clinical assessment methods, lab technology and radiography",

author = "Skregelid, \{Martine R.\} and Kasim Ibrahim-pur and Flemming Skj{\o}th and Pedersen, \{Malene Roland\} and Helle Precht",

year = "2026",

month = jan,

doi = "10.3390/radiation6010004",

language = "English",

volume = "6",

journal = "Radiation",

issn = "2673-592X",

publisher = "Multidisciplinary Digital Publishing Institute (MDPI)",

number = "1",

}

TY - JOUR

T1 - AI in Diagnostic Radiology: What Happens When Algorithms Are Updated

AU - Skregelid, Martine R.

AU - Ibrahim-pur, Kasim

AU - Skjøth, Flemming

AU - Pedersen, Malene Roland

AU - Precht, Helle

PY - 2026/1

Y1 - 2026/1

N2 - Hand and ankle fractures are common and can easily be missed in busy emergency departments. Therefore, many hospitals now use artificial intelligence (AI) to highlight possible fractures. AI software is frequently updated, and regular evaluation is warranted to assesswhether new updates alter the diagnostic performance. In this study, we investigated the diagnostic accuracy of two different versions of the same AI fracture detection software on hand and ankle radiographs, using reporting radiographers’ reports as the reference standard. We also assessed whether AI output led to changes in reporting radiographers’ assessments in cases of disagreement.

AB - Hand and ankle fractures are common and can easily be missed in busy emergency departments. Therefore, many hospitals now use artificial intelligence (AI) to highlight possible fractures. AI software is frequently updated, and regular evaluation is warranted to assesswhether new updates alter the diagnostic performance. In this study, we investigated the diagnostic accuracy of two different versions of the same AI fracture detection software on hand and ankle radiographs, using reporting radiographers’ reports as the reference standard. We also assessed whether AI output led to changes in reporting radiographers’ assessments in cases of disagreement.

KW - clinical assessment methods, lab technology and radiography

U2 - 10.3390/radiation6010004

DO - 10.3390/radiation6010004

M3 - Journal article

SN - 2673-592X

VL - 6

JO - Radiation

JF - Radiation

IS - 1

M1 - 4

ER -