Acceptance, Perception, and Technostress of Generative AI Among Radiologists at the University Hospital “Paolo Giaccone” of Palermo

Abstract

Introduction

The widespread adoption of generative artificial intelligence (AI) tools across multiple areas of medical practice, together with their predictive, diagnostic, and therapeutic applications, has raised growing interest in understanding healthcare professionals’ acceptance, use, and perceptions of this technology. At the same time, evidence suggests that AI implementation, particularly in the absence of adequate competencies, may contribute to technology-related stress (technostress).

Aims

This study aimed to investigate the mechanism through which AI perceptions influence technostress through the mediating role of AI acceptance among radiologists, a professional group particularly exposed to AI applications in clinical practice. In addition, the study examined whether self-efficacy, defined as individuals’ beliefs in their capabilities to execute behaviors necessary to achieve specific performance outcomes (Bandura, 1982), moderates the relationship between AI perceptions and acceptance.

Statistical methods

A sample of 71 physicians working at the Department of Radiology and Diagnostic Imaging of the University Hospital “Paolo Giaccone” of Palermo completed an online questionnaire administered through Google Forms. The survey included the AI Perception Scale (Shinners et al., 2022), the Acceptance Scale (Venkatesh et al., 2012), the Self-Efficacy Scale (Venkatesh et al., 2003), and the Technostress Scale (Tarafdar et al., 2007), which assesses overall technostress, techno-overload, techno-complexity/insecurity, and techno-uncertainty.

For each construct, exploratory factor analysis (EFA) was conducted to identify the underlying factor structure, followed by confirmatory factor analysis (CFA) to evaluate the fit of the measurement model against competing alternatives. Model fit was assessed using the Chi-square/df ratio (≤3), Root Mean Square Error of Approximation (RMSEA ≤0.08), Comparative Fit Index (CFI ≥0.90), Tucker–Lewis Index (TLI ≥0.90), and Standardized Root Mean Square Residual (SRMR ≤0.08). Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to compare competing models, with lower values indicating better fit. Conditional indirect effects of Perceptions of AI on technostress outcomes through AI Acceptance, were estimated at low (−1 SD), mean, and high (+1 SD) levels of Self-Efficacy, using the moderated mediation model implemented through the PROCESS macro Model 7 (Hayes, 2022) with 5000 bootstrap samples and bias-corrected 95% confidence intervals.

Results

AI perception was directly and positively associated with overall technostress (b = 0.57, 95% CI = 0.21–0.93), techno-overload (b = 0.58, 95% CI = 0.16–1.00), and techno-complexity/insecurity (b = 0.83, 95% CI = 0.42–1.23), but not with techno-uncertainty (b = −0.02, 95% CI = −0.46–0.43). AI acceptance was directly and negatively associated with the same outcomes (e.g., overall technostress: b = −0.55, 95% CI = −0.91 to −0.18). Furthermore, significant indirect effects of AI perception on technostress through AI acceptance emerged at both low (b = −0.36, 95% CI = −0.65 to −0.11) and moderate (b = −0.28, 95% CI = −0.53 to −0.09) levels of self-efficacy.

Conclusion

The findings suggest that AI perceptions influence technostress both directly and indirectly through AI acceptance. Therefore, the implementation of AI as a support tool for radiologists in routine clinical practice should aim to mitigate technostress by promoting AI acceptance, strengthening professional competencies, and addressing workload demands and organizational clarity.