Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance

Thomas Dratsch; Xue Chen; Mohammad Rezazade Mehrizi; Roman Kloeckner; Aline Mähringer-Kunz; Michael Püsken; Bettina Baeßler; Stephanie Sauer; David Maintz; Daniel Pinto dos Santos

doi:10.1148/radiol.222176

Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance

Thomas Dratsch^*, Xue Chen, Mohammad Rezazade Mehrizi, Roman Kloeckner, Aline Mähringer-Kunz, Michael Püsken, Bettina Baeßler, Stephanie Sauer, David Maintz, Daniel Pinto dos Santos

^*Corresponding author for this work

Research output: Contribution to Journal › Article › Academic › peer-review

179 Downloads (Pure)

Abstract

Background: Automation bias (the propensity for humans to favor suggestions from automated decision-making systems) is a known source of error in human-machine interactions, but its implications regarding artificial intelligence (AI)–aided mammography reading are unknown. Purpose: To determine how automation bias can affect inexperienced, moderately experienced, and very experienced radiologists when reading mammograms with the aid of an artificial intelligence (AI) system. Materials and Methods: In this prospective experiment, 27 radiologists read 50 mammograms and provided their Breast Imaging Reporting and Data System (BI-RADS) assessment assisted by a purported AI system. Mammograms were obtained between January 2017 and December 2019 and were presented in two randomized sets. The first was a training set of 10 mammograms, with the correct BI-RADS category suggested by the AI system. The second was a set of 40 mammograms in which an incorrect BI-RADS category was suggested for 12 mammograms. Reader performance, degree of bias in BI-RADS scoring, perceived accuracy of the AI system, and reader confidence in their own BI-RADS ratings were assessed using analysis of variance (ANOVA) and repeated-measures ANOVA followed by post hoc tests and Kruskal-Wallis tests followed by the Dunn post hoc test. Results: The percentage of correctly rated mammograms by inexperienced (mean, 79.7% ± 11.7 [SD] vs 19.8% ± 14.0; P <.001; r = 0.93), moderately experienced (mean, 81.3% ± 10.1 vs 24.8% ± 11.6; P <.001; r = 0.96), and very experienced (mean, 82.3% ± 4.2 vs 45.5% ± 9.1; P =.003; r = 0.97) radiologists was significantly impacted by the correctness of the AI prediction of BI-RADS category. Inexperienced radiologists were significantly more likely to follow the suggestions of the purported AI when it incorrectly suggested a higher BI-RADS category than the actual ground truth compared with both moderately (mean degree of bias, 4.0 ± 1.8 vs 2.4 ± 1.5; P =.044; r = 0.46) and very (mean degree of bias, 4.0 ± 1.8 vs 1.2 ± 0.8; P =.009; r = 0.65) experienced readers. Conclusion: The results show that inexperienced, moderately experienced, and very experienced radiologists reading mammograms are prone to automation bias when being supported by an AI-Based system. This and other effects of human and machine interaction must be considered to ensure safe deployment and accurate diagnostic performance when combining human readers and AI.

Original language	English
Article number	e222176
Pages (from-to)	1-9
Number of pages	9
Journal	Radiology
Volume	307
Issue number	4
Early online date	2 May 2023
DOIs	https://doi.org/10.1148/radiol.222176
Publication status	Published - May 2023

Bibliographical note

Funding Information:
From the Institute of Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne, Kerpener Str 62, 50937 Cologne, Germany (T.D., X.C., M.P., D.M., D.P.d.S.); School of Business and Economics, Knowledge, Information and Innovation, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands (M.R.M.); Institute of Interventional Radiology, University Clinic Schleswig-Holstein, Kiel, Germany (R.K.); Department of Diagnostic and Interventional Radiology, University Medical Centre of the Johannes Gutenberg-University Mainz, Mainz, Germany (A.M.K.); and Institute of Diagnostic and Interventional Radiology, University Clinic Würzburg, Würzburg, Germany (B.B., S.S.). Received September 9, 2022; revision requested November 2; revision received January 25, 2023; accepted March 13. Address correspondence to T.D. (email: [email protected]). Supported by a grant from the German Ministry of Health (BMG) as part of the project EVA-KI (ZMVI1-2520DAT03B). * T.D. and X.C. contributed equally to this work. Conflicts of interest are listed at the end of this article. See also the editorial by Baltzer in this issue.

Funding Information:
Disclosures of conflicts of interest: T.D. Grant from German Federal Ministry of Health. X.C. No relevant relationships. M.R.M. Patent held in association with Vrij University Amsterdam. R.K. Consulting fees from Boston Scientific, Bristol-Myers Squibb, Guerbet, Roche, and SIRTEX; payment or honoraria from BTG, EISAI, Guerbet, Ipsen, Roche, Siemens, SIRTEX, and MSD Sharp & Dohme. A.M.K. No relevant relationships. M.P. Honoraria for lectures from Bayer and Becton Dickinson; Bayer advisory board. B.B. Honorarium from Bayer Vital; founder and CEO of Lernrad. S.S. No relevant relationships. D.M. Honorarium from Philips. D.P.d.S. Grant from the German Federal Ministry of Health; consulting fees from Cook Medical; honorarium from Bayer; unpaid roles as Chair of the IT subcommittee of the German Radiological Society and Vice President of the European Society for Medical Imaging Informatics.

Publisher Copyright:
© RSNA, 2023.

Funding

From the Institute of Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne, Kerpener Str 62, 50937 Cologne, Germany (T.D., X.C., M.P., D.M., D.P.d.S.); School of Business and Economics, Knowledge, Information and Innovation, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands (M.R.M.); Institute of Interventional Radiology, University Clinic Schleswig-Holstein, Kiel, Germany (R.K.); Department of Diagnostic and Interventional Radiology, University Medical Centre of the Johannes Gutenberg-University Mainz, Mainz, Germany (A.M.K.); and Institute of Diagnostic and Interventional Radiology, University Clinic Würzburg, Würzburg, Germany (B.B., S.S.). Received September 9, 2022; revision requested November 2; revision received January 25, 2023; accepted March 13. Address correspondence to T.D. (email: [email protected]). Supported by a grant from the German Ministry of Health (BMG) as part of the project EVA-KI (ZMVI1-2520DAT03B). * T.D. and X.C. contributed equally to this work. Conflicts of interest are listed at the end of this article. See also the editorial by Baltzer in this issue. Disclosures of conflicts of interest: T.D. Grant from German Federal Ministry of Health. X.C. No relevant relationships. M.R.M. Patent held in association with Vrij University Amsterdam. R.K. Consulting fees from Boston Scientific, Bristol-Myers Squibb, Guerbet, Roche, and SIRTEX; payment or honoraria from BTG, EISAI, Guerbet, Ipsen, Roche, Siemens, SIRTEX, and MSD Sharp & Dohme. A.M.K. No relevant relationships. M.P. Honoraria for lectures from Bayer and Becton Dickinson; Bayer advisory board. B.B. Honorarium from Bayer Vital; founder and CEO of Lernrad. S.S. No relevant relationships. D.M. Honorarium from Philips. D.P.d.S. Grant from the German Federal Ministry of Health; consulting fees from Cook Medical; honorarium from Bayer; unpaid roles as Chair of the IT subcommittee of the German Radiological Society and Vice President of the European Society for Medical Imaging Informatics.

Funders	Funder number
European Society for Medical Imaging Informatics
German Ministry of Health
German Radiological Society
Bundesministerium für Gesundheit	ZMVI1-2520DAT03B
Bundesministerium für Gesundheit

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1148/radiol.222176

Dratsch et al. 2023 - Automation Bias in Mammography - The Impact of Artificial Intelligence BI-RADS Suggestions on Reader PerformanceFinal published version, 903 KBLicence: Article 25fa Dutch Copyright Act

Persistent URL (handle)

Copy link address

Cite this

@article{b6042501c3264c8e8b8b254042d11aff,

title = "Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance",

abstract = "Background: Automation bias (the propensity for humans to favor suggestions from automated decision-making systems) is a known source of error in human-machine interactions, but its implications regarding artificial intelligence (AI)–aided mammography reading are unknown. Purpose: To determine how automation bias can affect inexperienced, moderately experienced, and very experienced radiologists when reading mammograms with the aid of an artificial intelligence (AI) system. Materials and Methods: In this prospective experiment, 27 radiologists read 50 mammograms and provided their Breast Imaging Reporting and Data System (BI-RADS) assessment assisted by a purported AI system. Mammograms were obtained between January 2017 and December 2019 and were presented in two randomized sets. The first was a training set of 10 mammograms, with the correct BI-RADS category suggested by the AI system. The second was a set of 40 mammograms in which an incorrect BI-RADS category was suggested for 12 mammograms. Reader performance, degree of bias in BI-RADS scoring, perceived accuracy of the AI system, and reader confidence in their own BI-RADS ratings were assessed using analysis of variance (ANOVA) and repeated-measures ANOVA followed by post hoc tests and Kruskal-Wallis tests followed by the Dunn post hoc test. Results: The percentage of correctly rated mammograms by inexperienced (mean, 79.7% ± 11.7 [SD] vs 19.8% ± 14.0; P <.001; r = 0.93), moderately experienced (mean, 81.3% ± 10.1 vs 24.8% ± 11.6; P <.001; r = 0.96), and very experienced (mean, 82.3% ± 4.2 vs 45.5% ± 9.1; P =.003; r = 0.97) radiologists was significantly impacted by the correctness of the AI prediction of BI-RADS category. Inexperienced radiologists were significantly more likely to follow the suggestions of the purported AI when it incorrectly suggested a higher BI-RADS category than the actual ground truth compared with both moderately (mean degree of bias, 4.0 ± 1.8 vs 2.4 ± 1.5; P =.044; r = 0.46) and very (mean degree of bias, 4.0 ± 1.8 vs 1.2 ± 0.8; P =.009; r = 0.65) experienced readers. Conclusion: The results show that inexperienced, moderately experienced, and very experienced radiologists reading mammograms are prone to automation bias when being supported by an AI-Based system. This and other effects of human and machine interaction must be considered to ensure safe deployment and accurate diagnostic performance when combining human readers and AI.",

author = "Thomas Dratsch and Xue Chen and Mehrizi, {Mohammad Rezazade} and Roman Kloeckner and Aline M{\"a}hringer-Kunz and Michael P{\"u}sken and Bettina Bae{\ss}ler and Stephanie Sauer and David Maintz and {dos Santos}, {Daniel Pinto}",

note = "Funding Information: From the Institute of Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne, Kerpener Str 62, 50937 Cologne, Germany (T.D., X.C., M.P., D.M., D.P.d.S.); School of Business and Economics, Knowledge, Information and Innovation, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands (M.R.M.); Institute of Interventional Radiology, University Clinic Schleswig-Holstein, Kiel, Germany (R.K.); Department of Diagnostic and Interventional Radiology, University Medical Centre of the Johannes Gutenberg-University Mainz, Mainz, Germany (A.M.K.); and Institute of Diagnostic and Interventional Radiology, University Clinic W{\"u}rzburg, W{\"u}rzburg, Germany (B.B., S.S.). Received September 9, 2022; revision requested November 2; revision received January 25, 2023; accepted March 13. Address correspondence to T.D. (email: [email protected]). Supported by a grant from the German Ministry of Health (BMG) as part of the project EVA-KI (ZMVI1-2520DAT03B). * T.D. and X.C. contributed equally to this work. Conflicts of interest are listed at the end of this article. See also the editorial by Baltzer in this issue. Funding Information: Disclosures of conflicts of interest: T.D. Grant from German Federal Ministry of Health. X.C. No relevant relationships. M.R.M. Patent held in association with Vrij University Amsterdam. R.K. Consulting fees from Boston Scientific, Bristol-Myers Squibb, Guerbet, Roche, and SIRTEX; payment or honoraria from BTG, EISAI, Guerbet, Ipsen, Roche, Siemens, SIRTEX, and MSD Sharp & Dohme. A.M.K. No relevant relationships. M.P. Honoraria for lectures from Bayer and Becton Dickinson; Bayer advisory board. B.B. Honorarium from Bayer Vital; founder and CEO of Lernrad. S.S. No relevant relationships. D.M. Honorarium from Philips. D.P.d.S. Grant from the German Federal Ministry of Health; consulting fees from Cook Medical; honorarium from Bayer; unpaid roles as Chair of the IT subcommittee of the German Radiological Society and Vice President of the European Society for Medical Imaging Informatics. Publisher Copyright: {\textcopyright} RSNA, 2023.",

year = "2023",

month = may,

doi = "10.1148/radiol.222176",

language = "English",

volume = "307",

pages = "1--9",

journal = "Radiology",

issn = "0033-8419",

publisher = "Radiological Society of North America Inc.",

number = "4",

}

TY - JOUR

T1 - Automation Bias in Mammography

T2 - The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance

AU - Dratsch, Thomas

AU - Chen, Xue

AU - Mehrizi, Mohammad Rezazade

AU - Kloeckner, Roman

AU - Mähringer-Kunz, Aline

AU - Püsken, Michael

AU - Baeßler, Bettina

AU - Sauer, Stephanie

AU - Maintz, David

AU - dos Santos, Daniel Pinto

N1 - Funding Information: From the Institute of Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne, Kerpener Str 62, 50937 Cologne, Germany (T.D., X.C., M.P., D.M., D.P.d.S.); School of Business and Economics, Knowledge, Information and Innovation, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands (M.R.M.); Institute of Interventional Radiology, University Clinic Schleswig-Holstein, Kiel, Germany (R.K.); Department of Diagnostic and Interventional Radiology, University Medical Centre of the Johannes Gutenberg-University Mainz, Mainz, Germany (A.M.K.); and Institute of Diagnostic and Interventional Radiology, University Clinic Würzburg, Würzburg, Germany (B.B., S.S.). Received September 9, 2022; revision requested November 2; revision received January 25, 2023; accepted March 13. Address correspondence to T.D. (email: [email protected]). Supported by a grant from the German Ministry of Health (BMG) as part of the project EVA-KI (ZMVI1-2520DAT03B). * T.D. and X.C. contributed equally to this work. Conflicts of interest are listed at the end of this article. See also the editorial by Baltzer in this issue. Funding Information: Disclosures of conflicts of interest: T.D. Grant from German Federal Ministry of Health. X.C. No relevant relationships. M.R.M. Patent held in association with Vrij University Amsterdam. R.K. Consulting fees from Boston Scientific, Bristol-Myers Squibb, Guerbet, Roche, and SIRTEX; payment or honoraria from BTG, EISAI, Guerbet, Ipsen, Roche, Siemens, SIRTEX, and MSD Sharp & Dohme. A.M.K. No relevant relationships. M.P. Honoraria for lectures from Bayer and Becton Dickinson; Bayer advisory board. B.B. Honorarium from Bayer Vital; founder and CEO of Lernrad. S.S. No relevant relationships. D.M. Honorarium from Philips. D.P.d.S. Grant from the German Federal Ministry of Health; consulting fees from Cook Medical; honorarium from Bayer; unpaid roles as Chair of the IT subcommittee of the German Radiological Society and Vice President of the European Society for Medical Imaging Informatics. Publisher Copyright: © RSNA, 2023.

PY - 2023/5

Y1 - 2023/5

N2 - Background: Automation bias (the propensity for humans to favor suggestions from automated decision-making systems) is a known source of error in human-machine interactions, but its implications regarding artificial intelligence (AI)–aided mammography reading are unknown. Purpose: To determine how automation bias can affect inexperienced, moderately experienced, and very experienced radiologists when reading mammograms with the aid of an artificial intelligence (AI) system. Materials and Methods: In this prospective experiment, 27 radiologists read 50 mammograms and provided their Breast Imaging Reporting and Data System (BI-RADS) assessment assisted by a purported AI system. Mammograms were obtained between January 2017 and December 2019 and were presented in two randomized sets. The first was a training set of 10 mammograms, with the correct BI-RADS category suggested by the AI system. The second was a set of 40 mammograms in which an incorrect BI-RADS category was suggested for 12 mammograms. Reader performance, degree of bias in BI-RADS scoring, perceived accuracy of the AI system, and reader confidence in their own BI-RADS ratings were assessed using analysis of variance (ANOVA) and repeated-measures ANOVA followed by post hoc tests and Kruskal-Wallis tests followed by the Dunn post hoc test. Results: The percentage of correctly rated mammograms by inexperienced (mean, 79.7% ± 11.7 [SD] vs 19.8% ± 14.0; P <.001; r = 0.93), moderately experienced (mean, 81.3% ± 10.1 vs 24.8% ± 11.6; P <.001; r = 0.96), and very experienced (mean, 82.3% ± 4.2 vs 45.5% ± 9.1; P =.003; r = 0.97) radiologists was significantly impacted by the correctness of the AI prediction of BI-RADS category. Inexperienced radiologists were significantly more likely to follow the suggestions of the purported AI when it incorrectly suggested a higher BI-RADS category than the actual ground truth compared with both moderately (mean degree of bias, 4.0 ± 1.8 vs 2.4 ± 1.5; P =.044; r = 0.46) and very (mean degree of bias, 4.0 ± 1.8 vs 1.2 ± 0.8; P =.009; r = 0.65) experienced readers. Conclusion: The results show that inexperienced, moderately experienced, and very experienced radiologists reading mammograms are prone to automation bias when being supported by an AI-Based system. This and other effects of human and machine interaction must be considered to ensure safe deployment and accurate diagnostic performance when combining human readers and AI.

AB - Background: Automation bias (the propensity for humans to favor suggestions from automated decision-making systems) is a known source of error in human-machine interactions, but its implications regarding artificial intelligence (AI)–aided mammography reading are unknown. Purpose: To determine how automation bias can affect inexperienced, moderately experienced, and very experienced radiologists when reading mammograms with the aid of an artificial intelligence (AI) system. Materials and Methods: In this prospective experiment, 27 radiologists read 50 mammograms and provided their Breast Imaging Reporting and Data System (BI-RADS) assessment assisted by a purported AI system. Mammograms were obtained between January 2017 and December 2019 and were presented in two randomized sets. The first was a training set of 10 mammograms, with the correct BI-RADS category suggested by the AI system. The second was a set of 40 mammograms in which an incorrect BI-RADS category was suggested for 12 mammograms. Reader performance, degree of bias in BI-RADS scoring, perceived accuracy of the AI system, and reader confidence in their own BI-RADS ratings were assessed using analysis of variance (ANOVA) and repeated-measures ANOVA followed by post hoc tests and Kruskal-Wallis tests followed by the Dunn post hoc test. Results: The percentage of correctly rated mammograms by inexperienced (mean, 79.7% ± 11.7 [SD] vs 19.8% ± 14.0; P <.001; r = 0.93), moderately experienced (mean, 81.3% ± 10.1 vs 24.8% ± 11.6; P <.001; r = 0.96), and very experienced (mean, 82.3% ± 4.2 vs 45.5% ± 9.1; P =.003; r = 0.97) radiologists was significantly impacted by the correctness of the AI prediction of BI-RADS category. Inexperienced radiologists were significantly more likely to follow the suggestions of the purported AI when it incorrectly suggested a higher BI-RADS category than the actual ground truth compared with both moderately (mean degree of bias, 4.0 ± 1.8 vs 2.4 ± 1.5; P =.044; r = 0.46) and very (mean degree of bias, 4.0 ± 1.8 vs 1.2 ± 0.8; P =.009; r = 0.65) experienced readers. Conclusion: The results show that inexperienced, moderately experienced, and very experienced radiologists reading mammograms are prone to automation bias when being supported by an AI-Based system. This and other effects of human and machine interaction must be considered to ensure safe deployment and accurate diagnostic performance when combining human readers and AI.

UR - http://www.scopus.com/inward/record.url?scp=85159779314&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85159779314&partnerID=8YFLogxK

U2 - 10.1148/radiol.222176

DO - 10.1148/radiol.222176

M3 - Article

C2 - 37129490

AN - SCOPUS:85159779314

SN - 0033-8419

VL - 307

SP - 1

EP - 9

JO - Radiology

JF - Radiology

IS - 4

M1 - e222176

ER -

Automation Bias in Mammography: The Impact of Artificial Intelligence BI-RADS Suggestions on Reader Performance

Abstract

Bibliographical note

Funding

UN SDGs

Access to Document

Persistent URL (handle)

Other files and links

Fingerprint

Cite this