The Associations of Federated Machine Learning Algorithm's Perceived Trustworthiness and Robustness on User's Adoption Intent of AI-Based Tools in Radiology Medical Imaging
Purpose: To test the hypothesis whether user perception of trustworthiness and robustness of a machine learning algorithm using Federal machine learning approach of AI-based tools in radiology medical imaging has positive co-relation with user adoption intent.
Methodology: The quantitative study used perceived trustworthiness and robustness as the two independent variables and user adoption intent of AI-based tool in radiology medical imaging as the dependent variable. An online survey was planned as data collection instrument. The online survey consists of 12 research questionnaires, few of them on
Likert scale. To improve the quality and efficiency of the study, the Survey questionnaire was reviewed by experts and minor changes adopted. Pearson correlation was calculated on each independent variable versus the dependent variable and a linear regression was performed to test the correlation between both independent variables and the dependent variable.
Results: The survey link was sent to 256 recipients out of which 53 responded, giving the survey a 20.71% response rate, with 51 fully completing the surveys. Most of the respondents were Radiologists (45.10%), predominantly in the range of 10-15 years of experience (39.22%) and majority (35.29%) lived in India. Of 51 respondents, 44 (86.3%) had knowledge of federated learning and 7 (13.7%) had no knowledge of federated learning. Out of the 44 respondents who had knowledge of federated learning, 23 (52.27%) had moderate level of knowledge, 8 respondents had high level of knowledge, 1 respondent had very high and 5 (9.8%) had very low. Out of 51 respondents, 21 (41.2%) were currently a user of AI-based tools for radiology workflows and 22 (43.1%) either particpated or contributed in research or experiments related to federated learning or they intent to do. Due to the effect size (r=.549 for perceived trustworthiness and r=.303 for perceived robustness), it can also be stated that there is a moderately positive effect (medium) for trustworthiness and positively small effect for robustness individually in correlation with adoption intentions. The results from linear regression showed that model had adjusted r- squared of .302 indicating positive relationship. The average of the trustworthiness variable had a positive unstandardized beta of .586 and the average of Robustness had a positive unstandardized beta of .020.
In addition, an analysis of the experience ranges showed a potential difference in perceived adoption intentions in respondents in range of 10-15 years. The respondents with moderate and high level of knowledge had same mean for adoption intent (4.2). The respondents who were radiologists have relatively lower level of perception when compared with other respondents.
Conclusion: The independent variables - User perception of federated learning’s trustworthiness and robustness are statistically significant and influence positive correlation with dependent factor user adoption intent of AI-based tools in medical imaging in radiology. Also, it was concluded that difference in experience, levels of knowledge in federated learning and type of role of respondents had difference in answering the questionnaire and their perception. Further research could examine and provide valuable insights.