牙医对人工智能辅助决策支持用于口腔潜在恶性疾患(OPMD)检测的接受度:一项结构方程模型研究
Dentists’ acceptance of artificial intelligence-assisted decision support for oral potentially malignant disorder detection: a structural equation modeling study
一项结构方程模型研究评估牙医对人工智能辅助决策支持用于口腔潜在恶性疾患(OPMD)检测的接受度,模型在技术接受模型基础上纳入信任和对AI临床决策的依赖。研究关注影响牙医将此类AI工具纳入日常口腔诊疗的因素。
Introduction
Oral squamous cell carcinoma (OSCC) remains a significant global health burden, with more than 377,000 new cases and approximately 177,000 deaths annually according to GLOBOCAN 2020 estimates1. Despite advances in surgery, radiotherapy, chemotherapy, and targeted therapies, outcomes remain strongly influenced by the stage at diagnosis, with many patients continuing to present with advanced disease. Early diagnosis and appropriate staging are therefore important for improving treatment options, reducing treatment complexity, and optimizing patient outcomes1,2.
A proportion of oral squamous cell carcinomas (OSCCs) arise from clinically recognizable oral potentially malignant disorders (OPMDs), including leukoplakia, erythroplakia, oral lichen planus, and oral submucous fibrosis, which exhibit varying risks of malignant transformation. Early identification and appropriate management of these lesions provide a critical opportunity to prevent malignant transformation and improve patient outcomes3,4,5. However, distinguishing benign from potentially malignant oral lesions remains clinically challenging because early lesions often present with subtle or nonspecific clinical features. Moreover, oral mucosal lesions are frequently encountered in routine dental practice, emphasizing the importance of meticulous clinical examination and timely recognition of lesions requiring surveillance, biopsy, or specialist referral6. The clinical heterogeneity of OPMDs, including lesions such as oral lichen planus, further emphasizes the need for accurate risk assessment and appropriate surveillance7.
Dentists play a pivotal role in early detection, risk assessment, and referral of OPMDs because they are often the first healthcare professionals to encounter suspicious oral lesions during routine examinations. Nevertheless, deficiencies in oral cancer awareness and screening practices among dentists continue to be reported, highlighting the need for both educational initiatives and innovative clinical decision-support strategies to improve early detection and optimize referral pathways8.
Artificial intelligence (AI), particularly machine learning and deep learning, has emerged as a promising adjunct to clinical decision-making in oral medicine. AI systems have demonstrated promising diagnostic performance for detecting and classifying oral mucosal lesions using clinical images and multimodal data9,10,11. Systematic reviews, meta-analyses, and multicenter studies further support the potential of AI-assisted decision support for OPMD detection and oral cancer diagnosis across diverse clinical settings12,13,14,15,16. Despite these advances, translation into routine dental practice remains limited. Most published research has focused on algorithm development and diagnostic performance under controlled conditions, whereas comparatively little attention has been given to clinicians’ acceptance of AI-assisted decision support and the factors influencing its integration into routine dental practice14,15,16. Successful implementation depends not only on technical performance but also on clinicians’ willingness to trust and integrate AI into clinical workflows. Reported barriers include limited trust, inadequate explainability, workflow integration challenges, ethical and legal concerns, and insufficient education and training17,18,19,20,21. Similarly, Guo et al. (2026) identified clinician acceptance and implementation challenges as major barriers to the widespread adoption of AI for OPMD assessment15.
The Technology Acceptance Model (TAM) is one of the most widely validated frameworks for explaining technology acceptance and use22. According to TAM, perceived usefulness and perceived ease of use influence users’ attitudes and behavioral intention toward adopting new technologies22,23. Extensions of the model have demonstrated that incorporating constructs such as trust, perceived risk, and facilitating conditions improves its explanatory power, particularly in healthcare settings where technology directly influences clinical decision-making and patient care24,25,26,27. Trust is especially important for AI-assisted decision support because clinicians must critically evaluate AI-generated recommendations while retaining ultimate responsibility for clinical decision-making and patient care.
Although interest in AI in dentistry is increasing rapidly, evidence regarding dentists’ acceptance of AI-assisted decision support in oral medicine remains limited. Most previous studies have focused on awareness, educational readiness, or general perceptions of AI among dental students and practitioners28,29, whereas few have examined the determinants of dentists’ acceptance of AI-assisted decision support for OPMD detection using theory-driven implementation frameworks and structural equation modeling. Furthermore, evidence from low- and middle-income countries remains scarce despite the potential of AI to improve access to early detection and specialist referral where oral medicine expertise may be limited15,20.
Several theoretical frameworks have been developed to explain technology acceptance and adoption. The Unified Theory of Acceptance and Use of Technology (UTAUT) and its extension, UTAUT2, incorporate broader determinants such as social influence, facilitating conditions, habit, and other contextual factors30,31. TAM was selected for the present study because its parsimonious structure directly captures the cognitive pathway from perceived ease of use and perceived usefulness to attitude and behavioral intention22,23. This structure was considered particularly relevant to the present research question, which focused on dentists’ acceptance of AI-assisted clinical decision support for OPMD detection. The model was extended by incorporating trust and AI-assisted clinical decision reliance to capture factors specifically relevant to clinicians’ evaluation and potential reliance on AI recommendations24,25,26,27.
Few studies have applied a theory-driven framework to examine dentists’ acceptance of AI-assisted clinical decision support for OPMD detection, particularly in low- and middle-income countries. The present study addresses this gap by testing an extended Technology Acceptance Model that incorporates trust and a newly developed AI-Assisted Clinical Decision Reliance construct in a large sample of practicing dentists.
Accordingly, this cross-sectional study aimed to evaluate dentists’ acceptance of AI-assisted clinical decision support for oral potentially malignant disorder (OPMD) detection using an extended Technology Acceptance Model. Specifically, the study investigated the influence of perceived usefulness, perceived ease of use, trust in AI, attitude toward AI, and reliance on AI-assisted clinical decision support on behavioral intention to use AI-assisted decision support. Descriptive data on AI awareness, prior exposure, perceived implementation barriers, and educational needs were also collected.
Materials and methods
Study design and setting
This cross-sectional, questionnaire-based survey assessed Egyptian dentists’ acceptance of artificial intelligence (AI) for the early detection and clinical decision support for oral potentially malignant disorders (OPMDs). The study was guided by an extended Technology Acceptance Model (TAM) incorporating perceived ease of use, perceived usefulness, trust in AI, attitude toward AI, behavioral intention to use AI, and AI-assisted clinical decision reliance. The study was designed and reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
Participants
Licensed dentists currently practicing in Egypt were invited to participate through academic institutions, governmental healthcare facilities, private practices, and professional dental networks.
Inclusion criteria were licensed dentists currently practicing in Egypt, including general practitioners, specialists, consultants, academic staff members, and postgraduate residents, who provided informed consent.
Exclusion criteria included undergraduate dental students, retired or non-practicing dentists, duplicate responses, and questionnaires with substantial missing data.
Sample size calculation
The sample size was first estimated a priori using G*Power software version 3.1.9.7 for multiple linear regression analysis (fixed model, R2 deviation from zero)32. Because no previous Egyptian studies had investigated dentists’ acceptance of artificial intelligence for the early detection of oral potentially malignant disorders, a medium effect size (f2 = 0.15) was assumed according to Cohen’s recommendations33. With α = 0.05 and power = 0.80, the minimum required sample was 118 participants.
As the primary analysis involved confirmatory factor analysis and covariance-based structural equation modeling with six latent constructs and 25 observed indicators, the regression-based calculation was regarded only as a lower-bound estimate. A target sample of 500 dentists was therefore prespecified to support stable estimation of factor loadings and structural parameters, reliable standard errors, assessment of overall model fit, and the planned bootstrap mediation analyses in this relatively complex SEM34. The final analytic sample of 500 substantially exceeded the minimum regression-based requirement and met the planned sample size for the CFA and SEM analyses.
Questionnaire development
A structured self-administered questionnaire was developed by adapting validated Technology Acceptance Model (TAM) instruments22,24 together with previously published questionnaires evaluating artificial intelligence adoption in healthcare and dentistry26,28,29. The questionnaire was specifically adapted to evaluate dentists’ perceptions and acceptance of AI-assisted technologies for the early detection and clinical decision support for OPMDs.
The questionnaire consisted of 35 items organized into eight sections: (1) demographic and professional characteristics (6 items), (2) AI awareness and exposure (4 items), (3) Perceived Ease of Use (PEOU; 4 items), (4) Perceived Usefulness (PU; 4 items), (5) Trust in AI (4 items), (6) Attitude Toward AI (ATT; 4 items), (7) Behavioral Intention to Use AI (BI; 5 items), and (8) AI-Assisted Clinical Decision Reliance (AICDR; 4 items). The TAM constructs were included to assess key structural associations of technology acceptance, while the Trust construct was incorporated to capture confidence in the reliability and evidence base of AI recommendations. In addition, the AICDR construct was specifically developed to evaluate the extent to which AI recommendations may influence dentists’ clinical decision-making in situations involving suspicious oral lesions, thereby providing a clinically relevant extension beyond traditional measures of technology acceptance.
Items within Sections II–VIII were measured using a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree), with higher scores indicating greater awareness, acceptance, trust, intention to use AI, or reliance on AI-assisted clinical decision-making. The questionnaire was designed to be completed within approximately 8–10 min to maximize response rates while maintaining adequate coverage of the study constructs.
Content validation and pilot testing
The questionnaire underwent content validation by a multidisciplinary panel of five experts in oral medicine, oral pathology, periodontology, public health dentistry, and digital dentistry. Panel members assessed the items for relevance, clarity, and clinical appropriateness. Formal quantitative Content Validity Index (CVI) calculation was not performed. Feedback from the expert panel was used to refine item wording.
Following expert review, a pilot study involving 20 dentists was conducted to evaluate clarity, comprehensibility, and completion time. Feedback obtained during pilot testing was used to further refine item wording and questionnaire structure. Pilot responses were excluded from the final analysis.
Development of the AICDR scale
The AI-Assisted Clinical Decision Reliance (AICDR) construct was newly developed to extend the Technology Acceptance Model to AI-assisted OPMD decision-making. Items were developed by the research team based on the OPMD clinical decision-making context and literature on AI-assisted clinical decision support, reliance, and trust, addressing diagnostic confidence, referral, biopsy-related decisions, and reconsideration of clinical impressions.
No EFA or independent scale-development sample was used. The theoretically specified four-item construct was evaluated directly by CFA in the analytic sample (N = 500), showing standardized loadings of 0.72–0.92, AVE = 0.647, and CR = 0.879. These findings provide preliminary evidence of measurement reliability and convergent validity; further validation in independent samples is warranted.
Reliability and construct validation
The questionnaire comprised six latent constructs: Perceived Ease of Use (PEOU), Perceived Usefulness (PU), Trust in AI, Attitude Toward AI (ATT), Behavioral Intention to Use AI (BI), and AI-Assisted Clinical Decision Reliance (AICDR). The Trust construct assessed participants’ confidence in the reliability, consistency, and evidence base of AI-assisted recommendations. The AICDR construct was specifically developed to evaluate the extent to which AI recommendations influence dentists’ referral, diagnostic, and biopsy-related decisions in the management of suspicious oral lesions, thereby extending the traditional Technology Acceptance Model to a clinically relevant decision-making context.
Convergent validity was assessed using standardized factor loadings, average variance extracted (AVE), and composite reliability. Standardized factor loadings ≥ 0.70 were considered desirable, although loadings ≥ 0.50 were considered acceptable when supported by satisfactory composite reliability. Discriminant validity was evaluated using both the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT)35, with HTMT values < 0.85 indicating acceptable discriminant validity.
The four-item Perceived Ease of Use (PEOU) scale was adapted from established TAM instruments. One item (“Using AI would not significantly increase my clinical workload”) was retained on theoretical grounds because workload impact is repeatedly identified in the AI-in-healthcare literature as a critical practical barrier to the adoption of clinical decision-support systems. Although the item was conceptually relevant to dentists’ perceptions of ease of use in busy clinical settings, its psychometric performance was subsequently examined in a sensitivity analysis.
Data collection
Data were collected electronically using Google Forms. The survey link was distributed through professional dental networks, university mailing lists, and social media platforms. The first page of the survey contained study information and an electronic informed consent statement. Participation was voluntary and anonymous, and no personally identifiable information was collected. Responses were restricted to one submission per participant using Google Forms settings whenever technically feasible.
Recruitment and analytic sample
Licensed dentists practicing in Egypt were invited through academic institutions, governmental facilities, private practices, professional networks, university mailing lists, and social media. The survey was distributed electronically via Google Forms. Because the invitation was disseminated through multiple open networks, the precise number of individuals who accessed or started the questionnaire could not be retrieved, and a conventional response rate could not be calculated. After screening for eligibility, completeness, and response validity, 500 complete eligible responses were retained for analysis.
Ethical considerations
The study protocol was approved by the Research Ethics Committee of the Faculty of Dentistry, The British University in Egypt (Approval No. 26–040). Electronic informed consent was obtained from all participants before questionnaire completion. Participation was voluntary, and all responses were collected anonymously and handled confidentially in accordance with the Declaration of Helsinki.
Statistical analysis
Statistical analyses were performed using Python (version 3.10). Data management and descriptive analyses were conducted using the pandas, NumPy, and SciPy libraries. Descriptive statistics, including means, standard deviations, frequencies, and percentages, were used to summarize participant characteristics and questionnaire responses. Data completeness was assessed before analysis; all 500 included participants had complete responses for the 25 items comprising the six latent constructs, and therefore no missing-data imputation was required. Item-level distributional characteristics were examined using skewness and excess kurtosis. Observed skewness ranged from − 1.045 to − 0.004 and excess kurtosis from − 0.974 to 0.774.
Confirmatory factor analysis (CFA) and covariance-based structural equation modeling (CB-SEM) were conducted in Python 3.10 using the semopy structural equation modeling package (version 2.3.11). Maximum likelihood (ML) estimation was used, with questionnaire items treated as approximately continuous indicators. The six-factor measurement model comprised Perceived Ease of Use (PEOU), Perceived Usefulness (PU), Trust, Attitude Toward AI (ATT), Behavioral Intention to Use AI (BI), and AI-Assisted Clinical Decision Reliance (AICDR). The structural model specified PEOU as a predictor of PU; PU and Trust as predictors of ATT; and PEOU, PU, Trust, ATT, and AICDR as predictors of BI.
The measurement model was evaluated using standardized factor loadings, Cronbach’s α, composite reliability (CR), and average variance extracted (AVE). Standardized factor loadings of ≥ 0.70 were considered desirable, while loadings ≥ 0.50 were considered generally acceptable for established constructs. Items with lower loadings were not automatically removed solely based on this criterion when there was a clear theoretical rationale for their retention. Internal consistency was considered satisfactory for Cronbach’s α and CR values ≥ 0.70, while AVE ≥ 0.50 was considered evidence of adequate convergent validity. Discriminant validity was assessed using the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT).
Bivariate associations among the study constructs were examined using Pearson correlation coefficients. Pearson correlations were calculated from observed construct scores, whereas interconstruct correlations estimated in the CFA were latent-variable correlations; therefore, the two sets of coefficients were not expected to be numerically identical.
The hypothesized structural relationships were evaluated using CB-SEM based on the extended Technology Acceptance Model. Standardized path coefficients (β), standard errors, test statistics, and p-values were examined, together with coefficients of determination (R2) for endogenous constructs. Indirect effects were assessed using 5000 bootstrap resamples with 95% confidence intervals. The complete SEM model syntax and analysis code are provided in Supplementary File S5 to facilitate reproducibility of the analysis. An indirect effect was considered statistically significant when its 95% confidence interval did not include zero.
Overall model adequacy was evaluated using the χ2 statistic and degrees of freedom, χ2/df, Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA) with its confidence interval, and Standardized Root Mean Square Residual (SRMR).
Potential common-method variance was considered because the predictor and outcome constructs were assessed using a single self-administered questionnaire. Harman’s single-factor test was performed using an unrotated principal component analysis of the 25 measurement items. The proportion of variance accounted for by the first unrotated component was examined as a descriptive diagnostic rather than as a definitive test for the presence or absence of common-method variance.
All statistical tests were two-tailed, and statistical significance was set at p < 0.05.
Participant characteristics
A total of 500 licensed dentists practicing in Egypt participated in this study (Table 1). The mean age of the participants was 40.30 ± 12.82 years, and the sample was predominantly female (60.8%, n = 304). Most participants had more than 10 years of clinical experience (60.2%, n = 301). The most frequently represented specialties were periodontology (22.4%), general dentistry (19.0%), and oral medicine (17.2%). Participants primarily worked in academic (43.2%) or mixed (31.0%) practice settings. Only 26.0% (n = 130) had received formal training in oral cancer screening, whereas 74.0% (n = 370) reported no previous formal training (Table 1).
Descriptive statistics and reliability of the study constructs
Descriptive statistics and internal consistency estimates for all study constructs are presented in Table 2. Overall, participants reported moderate-to-high scores across all study constructs. The highest mean scores were observed for Attitude Toward AI (4.05 ± 0.88) and Behavioral Intention to Use AI (3.92 ± 0.85), followed by Perceived Usefulness (3.84 ± 0.92). Trust in AI and AI-Assisted Clinical Decision Reliance also showed moderate-to-high mean scores (3.68 ± 0.86 and 3.69 ± 0.86, respectively), whereas Perceived Ease of Use (PEOU) had the lowest mean score (3.52 ± 0.73) (Table 2).
All multi-item scales demonstrated Cronbach’s alpha values ranging from 0.662 for PEOU to 0.945 for Attitude Toward AI (Table 2). Item-level descriptive statistics and detailed reliability analyses, including corrected item–total correlations and Cronbach’s alpha if an item was deleted, are provided in Supplementary Tables S1 and S2.
Bivariate associations among study constructs
Pearson correlation analysis revealed statistically significant positive associations among all study constructs (all p < 0.001) (Table 3). Correlation coefficients ranged from 0.424 to 0.798, indicating moderate-to-strong positive relationships. Behavioral Intention to Use AI (BI) demonstrated the strongest correlation with Attitude Toward AI (r = 0.798), followed by Perceived Usefulness (r = 0.687) and Trust in AI (r = 0.688). Behavioral intention also showed a moderate positive association with Perceived Ease of Use (r = 0.538).
Among the antecedent constructs, PU was strongly correlated with ATT (r = 0.769) and Trust in AI (r = 0.758), while Trust in AI showed a similarly strong association with ATT (r = 0.785). PEOU demonstrated moderate positive correlations with ATT (r = 0.523), Trust in AI (r = 0.496), and BI (r = 0.538), and a comparatively weaker, although statistically significant, correlation with PU (r = 0.424). AICDR was positively correlated with all acceptance-related constructs, with the strongest associations observed with Trust in AI (r = 0.722), ATT (r = 0.721), and PU (r = 0.711), followed by BI (r = 0.667) and PEOU (r = 0.483). Overall, the correlation pattern was consistent with positive interrelationships among the proposed study constructs and provided support for proceeding to the hypothesized structural equation model.
Measurement model evaluation
The psychometric properties of the questionnaire were evaluated using confirmatory factor analysis (CFA). Standardized factor loadings were satisfactory for most indicators, although one PEOU item showed a substantially lower loading of 0.170. For PEOU, three items showed loadings of 0.742, 0.864, and 0.697, whereas the item “Using AI would not significantly increase my clinical workload” showed a loading of 0.170. We retained the low-loading item in the primary analysis because the four-item PEOU construct was theoretically prespecified; we then examined its psychometric contribution through sensitivity analysis.
Composite reliability (CR) values ranged from 0.736 to 0.944, while AVE ranged from 0.453 to 0.810 (Table 4). The PEOU construct showed marginal convergent validity, with AVE = 0.453, although its CR was 0.736. The remaining constructs demonstrate satisfactory CR and AVE values. The Fornell–Larcker criterion supported discriminant validity (Table 4) and HTMT, with all HTMT values ranging from 0.528 to 0.842, below the conservative threshold of 0.85 (Supplementary Table S3). Overall, the measurement model demonstrated generally satisfactory reliability and discriminant validity, although the PEOU construct showed a marginal AVE and one substantially low-loading item.
Structural equation modeling
Given the satisfactory reliability and validity of the measurement model, the hypothesized structural relationships were evaluated using structural equation modeling (SEM). As illustrated in Fig. 1, the structural model demonstrated a good overall fit to the observed data (χ2/df = 2.41, CFI = 0.962, TLI = 0.956, RMSEA = 0.053, and SRMR = 0.044).
All hypothesized structural paths in the primary four-item model were statistically significant (Table 5). Attitude toward AI showed the strongest direct association with behavioral intention (β = 0.550, p < 0.001), followed by AI-assisted clinical decision reliance (β = 0.220), perceived usefulness (β = 0.180), trust in AI (β = 0.140), and perceived ease of use (β = 0.120). Perceived ease of use was strongly associated with perceived usefulness (β = 0.620), while both perceived usefulness (β = 0.480) and trust in AI (β = 0.400) were significantly associated with attitude toward AI. The structural model explained 38.4% of the variance in perceived usefulness, 64.8% of the variance in attitude toward AI, and 67.5% of the variance in behavioral intention (Fig. 1).
Bootstrap mediation analysis identified significant indirect effects for the three prespecified indirect pathways (Table 6). Perceived Ease of Use exerted a significant indirect effect on Behavioral Intention through Perceived Usefulness and Attitude Toward AI (indirect β = 0.164, 95% CI = 0.118–0.214, p < 0.001). Trust in AI also demonstrated a significant indirect effect through Attitude Toward AI (indirect β = 0.220, 95% CI = 0.165–0.282, p < 0.001), as did Perceived Usefulness (indirect β = 0.264, 95% CI = 0.198–0.339, p < 0.001).
Structural equation model based on the extended Technology Acceptance Model (TAM) illustrating the significant relationships among the study constructs (N = 500). Standardized path coefficients (β) and coefficients of determination (R2) for Perceived Usefulness (0.384), Attitude Toward AI (0.648), and Behavioral Intention (0.675) are shown. Overall model fit: χ2/df = 2.41, CFI = 0.962, TLI = 0.956, RMSEA = 0.053, and SRMR = 0.044. All displayed structural paths in the primary four-item PEOU model were statistically significant (p < 0.05).
Assessment of common-method bias
Because all principal predictor and outcome constructs were measured using a single self-administered questionnaire completed by the same respondents at one time point and using the same response format, the potential for common-method variance was considered. Harman’s single-factor analysis showed that the first unrotated component accounted for 55.51% of the total variance, exceeding the commonly used 50% heuristic. This finding raises concern about potential common-method variance but does not by itself establish its presence or magnitude.
HTMT ratios were also examined. All values were below the conservative threshold of 0.85, although the Trust–Attitude toward AI association was relatively high (HTMT = 0.842), indicating substantial construct overlap that warrants cautious interpretation. Importantly, HTMT is primarily a test of discriminant validity rather than a direct test of common-method bias. Collectively, these findings indicate that common-method variance cannot be excluded and that some structural associations, particularly Trust → Attitude and Attitude → Behavioral Intention, should be interpreted cautiously.
Additional findings
Moderate awareness and prior exposure to AI technologies were observed among participants, with awareness item means generally ranging from 3.0 to 3.5 (Supplementary Table S1). Despite limited previous experience with AI-assisted diagnostic systems, respondents expressed a strong willingness to receive future AI-related training, with the highest agreement for the statement, “I intend to seek training to improve my ability to use AI in dentistry” (4.12 ± 0.89). Item-level responses also indicated that workflow integration and increased clinical workload were the most frequently reported barriers to AI implementation, whereas participants strongly agreed that AI could improve early detection of OPMDs, referral decisions, and clinical decision-making.
Discussion
Artificial intelligence (AI) has emerged as one of the most transformative technologies in healthcare, with increasing applications in oral medicine for the early detection, diagnosis, and clinical decision support of OPMDs9,10,11. Early identification of OPMDs remains a major clinical priority because delayed diagnosis substantially increases the risk of malignant transformation and contributes to poorer patient outcomes1,3. Recent advances in deep learning, computer vision, and AI-assisted clinical imaging have demonstrated considerable promise in improving the detection of suspicious oral lesions, supporting diagnostic decision-making, and facilitating timely referral12,13,16. Nevertheless, the successful implementation of AI in routine dental practice depends not only on technological performance but also on clinicians’ willingness to adopt, trust, and appropriately integrate these systems into clinical decision-making17,25,26. Against this background, the present study employed an extended Technology Acceptance Model (TAM) evaluated using covariance-based structural equation modeling. Importantly, AI should be regarded as a clinical decision-support tool that complements, rather than replaces, comprehensive clinical examination, professional judgment, and histopathological confirmation when indicated15,20.
The measurement model demonstrated adequate psychometric quality, providing a sound basis for interpreting the relationships among the proposed determinants of AI acceptance. Importantly, the evidence that the constructs were sufficiently distinct suggests that dentists differentiated between related but conceptually different perceptions of AI, including its ease of use, usefulness, trustworthiness, and their resulting attitudes and behavioral intentions. This distinction is particularly relevant in the context of clinical AI, where a system may be perceived as useful without necessarily being considered easy to use or sufficiently trustworthy for clinical decision-making.
Although Perceived Ease of Use showed a marginally sub-threshold average variance extracted, its composite reliability and overall measurement performance remained acceptable according to established SEM guidance34,36. This finding warrants some caution in interpreting the PEOU construct but does not undermine its inclusion in the model. More broadly, the satisfactory measurement properties and overall model fit indicate that the extended Technology Acceptance Model provides a coherent framework for examining dentists’ acceptance of AI-assisted decision support. Rather than reflecting acceptance as a single undifferentiated attitude, the model supports a more nuanced interpretation in which usability, perceived clinical value, trust, and reliance on AI-assisted decision support represent related but distinguishable dimensions of adoption37,38,39. This distinction is important because acceptance of clinical AI may depend not simply on whether dentists view technology favorably, but on whether they perceive it as usable, clinically valuable, trustworthy, and appropriate for incorporation into professional decision-making.
One of the principal findings was the overall positive attitude toward AI-assisted technologies. Participants reported high levels of perceived usefulness, favorable attitudes, and strong behavioral intentions despite only one-quarter having previously received formal training in oral cancer screening. These findings suggest that clinicians recognize the potential clinical value of AI even when practical experience remains limited. Similar observations have been reported internationally, where dentists generally express enthusiasm regarding AI-assisted diagnostics but simultaneously acknowledge deficiencies in education, practical exposure, and implementation guidance10,11,29,40. Likewise, a recent Egyptian study among postgraduate dental students demonstrated widespread interest in AI applications despite concerns about reliability, ethics, and insufficient formal education, underscoring the growing need to incorporate AI competencies into dental curricula28. The coexistence of high acceptance with limited formal training is noteworthy: it may reflect optimism about the potential of AI rather than experience-based confidence in its real-world performance. Accordingly, favorable attitudes should not be interpreted as evidence that dentists are already adequately prepared to use AI safely and critically. Structured education, hands-on exposure, and guidance on appropriate human oversight may therefore be necessary to translate positive perceptions into competent clinical use.
Consistent with the original Technology Acceptance Model, Perceived Ease of Use demonstrated a strong positive association with Perceived Usefulness. Dentists who perceived AI systems as easier to learn and integrate into clinical workflows were substantially more likely to recognize their clinical utility. This relationship has consistently been reported since the original TAM and its subsequent extensions and remains one of the most stable predictors of technology acceptance across healthcare environments22,23,24,25. Although Perceived Ease of Use also exhibited a statistically significant direct association with Behavioral Intention, its influence was considerably strengthened through indirect pathways involving Perceived Usefulness and Attitude Toward AI. Taken together, these findings suggest that usability is associated with greater AI adoption by enabling dentists to recognize how AI can be incorporated into existing diagnostic workflows without creating additional cognitive or operational burden. However, the association between PEOU and BI was weaker than its indirect contribution through PU and ATT, suggesting that usability alone may be insufficient to drive adoption when the clinical value of the technology is unclear. This finding suggests that intuitive system design is associated with greater adoption primarily by enabling clinicians to appreciate the practical clinical benefits of AI-assisted diagnosis rather than through usability alone, emphasizing that user-friendly systems are more readily adopted when they clearly enhance clinical decision-making.
These findings reinforce the central premise of the Technology Acceptance Model that clinicians are primarily motivated to adopt new technologies when they perceive clear improvements in diagnostic accuracy, efficiency, productivity, and patient care22,26,41. Within the context of OPMDs, this finding is particularly relevant because numerous AI systems have demonstrated promising diagnostic accuracy in identifying potentially malignant oral lesions from clinical photographs and other imaging modalities12,13,16,38. Recent systematic reviews and meta-analyses further support the diagnostic potential of AI-assisted detection of OPMDs and oral cancer12,13. However, perceived usefulness should be distinguished from demonstrated clinical effectiveness. Promising diagnostic performance in research settings does not necessarily establish that an AI system improves patient outcomes, diagnostic pathways, or clinical decision-making in routine practice. Since dentists routinely encounter a broad spectrum of oral mucosal lesions in daily practice, AI-assisted clinical decision support has the potential to improve the consistency of lesion assessment, facilitate timely surveillance, biopsy, and specialist referral, and ultimately strengthen oral cancer screening pathways, particularly for high-risk patients6. Future implementation research should therefore examine whether perceived usefulness translates into measurable improvements in diagnostic accuracy, referral appropriateness, workflow efficiency, and patient outcomes rather than assuming that positive perceptions alone will produce these benefits.
Trust in AI was significantly associated with attitude toward AI and showed a substantial indirect association with behavioral intention mediated through attitude. These findings are consistent with contemporary AI adoption frameworks, which emphasize that clinicians must perceive AI systems as reliable, transparent, and clinically dependable before incorporating their recommendations into patient care17,18,41. The importance of trust in the present study extends the conventional TAM perspective by indicating that perceived usefulness may not be sufficient when clinical decisions are involved. Dentists may recognize that an AI system is potentially useful while remaining reluctant to incorporate its recommendations if they cannot judge when or why those recommendations should be trusted. Explainable AI, transparent algorithm development, external validation, and continuous post-implementation monitoring are therefore likely to play fundamental roles in strengthening clinician confidence20,21. At the same time, trust should not be equated with unconditional reliance. Appropriate clinical implementation requires calibrated trust, whereby dentists understand both the capabilities and limitations of an AI system and retain responsibility for integrating its output with clinical examination and other diagnostic evidence. Ethical considerations, including accountability, fairness, privacy, and algorithmic bias, must similarly accompany technological development to ensure responsible implementation within clinical dentistry19,20,42.
A notable contribution of the present study is the incorporation of AI-Assisted Clinical Decision Reliance as an extension of the traditional Technology Acceptance Model. Unlike perceived usefulness or trust, this construct specifically evaluates clinicians’ willingness to integrate AI recommendations into real diagnostic decision-making while maintaining professional responsibility. AI-Assisted Clinical Decision Reliance showed an independent positive association with Behavioral Intention after controlling for all conventional TAM constructs. This distinction is particularly important for OPMD detection because the clinical value of AI ultimately depends on how its outputs are incorporated into decisions concerning further examination, referral, biopsy, or surveillance. This distinction appears to have received less attention in previous TAM-based investigations of healthcare AI adoption, which have primarily emphasized perceived usefulness, perceived ease of use, trust, and attitudes26,41. The present findings therefore extend conventional technology-acceptance models from general willingness to use AI toward the more clinically specific question of whether clinicians are prepared to incorporate AI-generated information into their diagnostic reasoning. As AI systems increasingly function as clinical decision-support tools rather than purely analytical software, clinicians’ confidence in integrating AI outputs into referral, biopsy, and surveillance decisions may become a key determinant of successful clinical implementation20,43.
Taken together, the findings suggest that dentists’ acceptance of AI-assisted OPMD decision support is best understood as a multidimensional process rather than as a direct response to technological capability alone. Perceived ease of use appears to contribute to recognition of clinical usefulness, while perceived usefulness and trust shape dentists’ overall evaluation of AI and, consequently, their intention to use it. The additional contribution of AI-Assisted Clinical Decision Reliance further suggests that acceptance of clinical AI involves a transition from recognizing potential benefits to considering whether AI-generated information can be appropriately incorporated into professional decision-making. This extended perspective is particularly relevant to oral medicine, where AI outputs may influence decisions with consequences for biopsy, referral, surveillance, and early cancer detection.
Despite these potential benefits, the clinical application of AI-assisted OPMD detection remains subject to important technical and clinical limitations. AI performance depends strongly on the quality, diversity, and representativeness of training and validation datasets, while differences in lesion prevalence, patient characteristics, imaging devices, acquisition conditions, and annotation practices may limit generalizability across populations and clinical settings. This is particularly relevant to OPMDs, which exhibit substantial clinical and morphological heterogeneity. AI systems may also generate false-positive or false-negative recommendations, particularly for subtle, borderline, or early lesions. False-negative outputs could contribute to delayed referral, biopsy, or diagnosis, whereas false-positive outputs could result in unnecessary investigations, referrals, patient anxiety, and healthcare utilization. Accordingly, AI outputs should be regarded as decision-support information rather than substitutes for comprehensive clinical examination, professional judgment, or histopathological assessment when indicated. Moreover, prospective evidence from routine clinical practice remains limited, and further evaluation is needed to establish whether AI-assisted OPMD detection improves diagnostic pathways and patient outcomes beyond performance demonstrated in retrospective or controlled datasets20,43,44,45.
Regulatory and medicolegal considerations also require attention before routine clinical implementation. Issues including accountability when AI outputs conflict with professional judgment, transparency of algorithmic decision-making, responsibility for errors, data governance, validation requirements, and post-deployment monitoring require clearly defined institutional and regulatory frameworks. Thus, clinician acceptance should not be equated with clinical readiness for implementation. Responsible adoption requires appropriately validated systems, transparent performance reporting, human oversight, and clearly defined professional and regulatory responsibilities20,41,44.
Attitude Toward AI showed the strongest direct association with Behavioral Intention and served as a significant mediator linking Perceived Usefulness and Trust with technology adoption. The mediation analysis demonstrated that favorable perceptions of AI are associated with higher behavioral intention largely through the development of positive attitudes toward AI rather than through direct effects alone. These findings are highly consistent with Technology Acceptance Model 3 and meta-analytic evidence demonstrating that attitude represents an important mechanism through which cognitive beliefs influence behavioral intentions24,27. Consequently, implementation strategies should aim not only to improve technical performance but also to foster positive clinician attitudes through education, demonstration of clinical benefits, and opportunities for practical experience.
The structural model explained 67.5% of the variance in Behavioral Intention, indicating substantial explanatory power. This level of explained variance compares favorably with many previous healthcare technology acceptance studies25,26,46,47. The relatively high explained variance observed in the present study may partly reflect the inclusion of Trust and AI-Assisted Clinical Decision Reliance, two constructs that capture important psychological dimensions specific to AI-supported clinical decision-making. These findings illustrate the potential value of extending classical TAM frameworks with constructs that capture trust and clinically specific reliance when studying AI-supported healthcare decision-making27,41.
From a clinical perspective, the findings have important implications for oral cancer screening and the future integration of AI-assisted diagnostic pathways into dental practice. AI-assisted analysis of intraoral photographs and other clinical data may support opportunistic oral cancer screening during routine dental examinations by improving lesion recognition, reducing inter-clinician variability, facilitating risk stratification, and promoting more consistent referral and biopsy decisions10,15,16. Because many patients initially present to general dental practitioners rather than oral medicine specialists, AI-based decision-support systems may enhance the early identification of high-risk lesions at the primary care level, shorten diagnostic pathways, and reduce delays in specialist referral and definitive diagnosis12,13,38.
The finding that workflow integration represented one of the principal perceived barriers emphasizes that successful AI implementation will depend not only on diagnostic accuracy but also on effective incorporation into existing clinical processes18,25,26,42. AI systems should be designed to integrate seamlessly with electronic dental records, digital imaging platforms, referral pathways, multidisciplinary oral cancer networks, and teleconsultation services without increasing consultation time or administrative burden. Such interoperability may facilitate real-time clinical decision support, particularly in underserved or geographically remote areas where specialist resources are less readily available15,20,43.
Despite favorable attitudes toward AI, participants demonstrated only moderate familiarity with AI technologies and limited formal training, highlighting an important gap between acceptance and implementation readiness. This finding is consistent with previous evidence indicating deficiencies in oral cancer awareness and screening preparedness among dentists and suggests that AI-assisted decision-support systems may help standardize lesion assessment and improve referral and biopsy decisions when appropriately integrated into clinical workflows8. Educational initiatives should prioritize AI literacy, critical interpretation of algorithmic outputs, ethical decision-making, and integration of AI into evidence-based clinical pathways. AI competencies should therefore be progressively incorporated into undergraduate and postgraduate dental education, while continuing professional development programs should provide supervised practical exposure to validated AI systems9,10,48. In parallel, collaboration among educators, professional organizations, policymakers, developers, and regulatory authorities will be essential to establish governance frameworks that promote the safe, transparent, explainable, and responsible use of AI in oral healthcare20,41,49.
To our knowledge, this is among the first studies to evaluate dentists’ acceptance of AI for OPMD detection using a rigorously evaluated covariance-based structural equation modeling framework and to examine AI-Assisted Clinical Decision Reliance as a distinct construct within an extended Technology Acceptance Model. The study provided a comprehensive assessment of measurement validity, reliability, direct and indirect associations, mediation pathways, and explained variance. Furthermore, the relatively large sample size enhances the methodological robustness of the findings and provides valuable insights into the factors influencing AI acceptance among dental professionals.
Nevertheless, several limitations should be acknowledged. First, the cross-sectional design limits causal inference. Although structural equation modeling was used to test associations consistent with the proposed theoretical model, the observed paths do not establish temporal precedence or causality. Longitudinal or experimental designs are required to determine whether the identified associations translate into sustained AI adoption. Second, the study assessed behavioral intention rather than actual clinical use. Accordingly, the R² of 0.675 indicates that the model explained 67.5% of the variance in respondents’ stated intention to use AI within this sample; it should not be interpreted as explaining actual clinical adoption or implementation. Behavioral intention may not translate directly into routine use because implementation is additionally influenced by system performance, workflow integration, training, cost, institutional policies, regulatory requirements, and professional accountability. Prospective studies incorporating objective measures of AI use are therefore required to determine whether the observed acceptance translates into sustained clinical implementation.
Third, most participants had limited direct exposure to AI-based diagnostic technologies, and their responses may therefore reflect anticipated rather than experience-based perceptions of AI adoption. The reliance on self-reported perceptions may also have introduced response bias. Because all principal constructs were measured using a self-administered questionnaire completed by the same respondents at a single time point and using the same response format, common-method variance cannot be excluded. Harman’s single-factor analysis showed that the first factor accounted for 55.51% of the total variance, exceeding the commonly used 50% heuristic. This finding raises concern about potential common-method variance but does not by itself establish its presence or magnitude. Although construct distinctiveness was supported by confirmatory factor analysis, the Fornell–Larcker criterion, and HTMT ratios, these procedures do not eliminate the possibility of shared method-related variance. The relatively high HTMT value between Trust and Attitude toward AI (0.842) further warrants cautious interpretation. Future studies should incorporate multi-source data, temporally separated assessment of predictors and outcomes, or objective measures of AI use to reduce this potential source of bias.
Fourth, one item of the Perceived Ease of Use scale demonstrated a low corrected item–total correlation and factor loading. Although sensitivity analysis showed improved measurement properties after its removal, the a priori four-item specification was retained as the primary model because of the item’s theoretical relevance. This finding represents a measurement limitation and warrants further validation of the construct in independent samples.
Fifth, the open electronic recruitment strategy precludes calculation of a conventional response rate and may have introduced selection bias, as dentists with greater interest in artificial intelligence or oral cancer screening may have been more likely to participate. Although participants represented multiple specialties and practice settings, the findings may not be directly generalizable to countries with different healthcare infrastructures, regulatory environments, or levels of AI implementation.
Finally, responsible clinical implementation requires transparency regarding AI involvement in patient assessment. Patients should be informed when AI contributes to their clinical assessment, with appropriate informed-consent processes in accordance with applicable ethical and institutional requirements. Future studies should therefore examine actual AI utilization, longitudinal changes in clinician acceptance, educational interventions, governance frameworks, and real-world clinical performance across diverse healthcare settings.
Conclusion
Egyptian dentists showed favorable behavioral intention toward AI-assisted decision support for OPMD detection and assessment. Perceived usefulness, trust, attitude, and reliance on AI-assisted clinical decisions were significantly associated with behavioral intention within the proposed TAM framework. These findings reflect stated behavioral intention rather than actual clinical use and do not establish causal relationships. Translation into practice will require externally validated and explainable AI systems, prospective real-world evaluation, clinician education, appropriate workflow integration, and clear regulatory and professional accountability. AI should currently be regarded as a supplementary decision-support tool under appropriate human clinical oversight, rather than as a replacement for professional judgment or a definitive diagnostic tool. Histopathological examination remains the reference standard when tissue diagnosis is indicated. Where AI contributes to patient assessment, appropriate transparency and informed consent should be maintained in accordance with applicable ethical, institutional, and regulatory requirements.
Sensitivity analysis of the perceived ease of use construct
The original four-item PEOU scale demonstrated a Cronbach’s α of 0.662. The item “Using AI would not significantly increase my clinical workload” showed a corrected item–total correlation of 0.185 and a standardized factor loading of 0.170; its deletion increased Cronbach’s α to 0.804.
Re-estimation of the measurement model after removing this item yielded standardized loadings of 0.740, 0.870, and 0.692, with AVE = 0.594 and CR = 0.813 (Supplementary Table S4), indicating improved internal consistency and convergent validity of the reduced PEOU construct.
The full structural model was subsequently re-estimated using the three-item PEOU specification. PEOU remained positively associated with perceived usefulness (β = 0.459, p < 0.001), with the model explaining 21.1% of the variance in perceived usefulness. The model explained 68.7% of the variance in attitude and 68.9% of the variance in behavioral intention. Several structural estimates differed from those of the primary model, including a stronger PEOU→BI association (β = 0.210, p < 0.001) and a nonsignificant Trust→BI association (β = 0.050, p = 0.271). Bootstrap analysis based on 5,000 resamples also supported significant indirect associations through the specified TAM pathways in the re-estimated model.
The four-item PEOU specification was retained as the primary analysis because it represented the prespecified theoretical construct, whereas item removal was prompted by examination of the observed psychometric results. The three-item specification is therefore presented as a sensitivity analysis rather than as a replacement for the primary model. The weak empirical contribution of the workload item is acknowledged as a measurement limitation, and the findings involving PEOU are interpreted accordingly.
References
Sung, H. et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J. Clin. 71 (3), 209–249. https://doi.org/10.3322/caac.21660 (2021).
Agrawal, R. et al. Early detection of oral cavity cancer: a comprehensive literature review of risk factors and latest techniques in diagnosis. Curr. Surg. Rep. 13, 15. https://doi.org/10.1007/s40137-024-00442-8 (2025).
Warnakulasuriya, S. et al. Oral potentially malignant disorders: a consensus report from an international seminar on nomenclature and classification, convened by the WHO Collaborating Centre for Oral Cancer. Oral Dis. 27 (8), 1862–1880. https://doi.org/10.1111/odi.13704 (2021).
Nabil, G. et al. Evaluation of the chemopreventive effect of thymoquinone on oral leukoplakia: a randomized clinical trial. Explor. Med. 6, 1001290. https://doi.org/10.37349/emed.2025.1001290 (2025).
Ghalwash, D., Ammar, A., Abou-Bakr, A., Diab, A. H. & El-Gawish, A. Validation of salivary proteomic biomarkers for early detection of oral cancer in the Egyptian population. Future Sci. OA. 11 (1). https://doi.org/10.1080/20565623.2024.2432222 (2025).
Ghalwash, D., Abd El-Wahed, S. & Hamdy, A. Prevalence of oral mucosal lesions among geriatric dental patients in Egypt: a cross-sectional study. J. Dent. Res. Dent. Clin. Dent. Prospects. 19 (3), 204–212. https://doi.org/10.34172/joddd.025.43929 (2025).
Zahran, F. M., Elsaadany, B., Azab, N., El-Gawish, A. & Ghalwash, D. Dysplasia in oral lichen planus in a sample of Egyptians attending two tertiary care centers in Cairo. Oral Dis. https://doi.org/10.1111/odi.15182 (2024).
Ghalwash, D. & Zahran, F. Oral cancer awareness among dentists: what is missing? A cross-sectional study. J. Egypt. Natl. Canc Inst. 37, 31. https://doi.org/10.1186/s43046-025-00290-2 (2025).
Chen, Y. W., Stanley, K. & Att, W. Artificial intelligence in dentistry: current applications and future perspectives. Quintessence Int. 51 (3), 248–257. https://doi.org/10.3290/j.qi.a43952 (2020). (Erratum in: Quintessence Int. 2020;51(5):430. doi: 10.3290/j.qi.a44465).
Schwendicke, F., Samek, W. & Krois, J. Artificial intelligence in dentistry: Chances and challenges. J. Dent. Res. 99 (7), 769–774. https://doi.org/10.1177/0022034520915714 (2020).
Khanagar, S. B. et al. Developments, application, and performance of artificial intelligence in dentistry—A systematic review. J. Dent. Sci. 16 (1), 508–522. https://doi.org/10.1016/j.jds.2020.06.019 (2021).
Li, J. et al. Diagnostic accuracy of artificial intelligence-assisted clinical imaging in the detection of oral potentially malignant disorders and oral cancer: a systematic review and meta-analysis. Int. J. Surg. 110 (8), 5034–5046. https://doi.org/10.1097/JS9.0000000000001469 (2024).
Sahoo, R. K. et al. Diagnostic performance of artificial intelligence in detecting oral potentially malignant disorders and oral cancer using medical diagnostic imaging: a systematic review and meta-analysis. Front. Oral Health. 5, 1494867. https://doi.org/10.3389/froh.2024.1494867 (2024).
Ye, Y. J., Han, Y., Liu, Y., Guo, Z. L. & Huang, M. W. Utilizing deep learning for automated detection of oral lesions: A multicenter study. Oral Oncol. 155, 106873. https://doi.org/10.1016/j.oraloncology.2024.106873 (2024).
Guo, X. et al. Application of artificial intelligence in oral potentially malignant disorders: current opinions and future barriers. Clin. Transl. Oncol. 28, 804–817. https://doi.org/10.1007/s12094-025-04043-4 (2026).
Zhao, X. et al. Recent advances in early oral lesion diagnosis: the application of artificial intelligence-assisted endoscopy. Front. Oncol. 15, 1686356. https://doi.org/10.3389/fonc.2025.1686356 (2026).
Amann, J., Blasimme, A., Vayena, E., Frey, D. & Madai, V. I. Precise4Q Consortium. Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Med. Inf. Decis. Mak. 20 (1), 310. https://doi.org/10.1186/s12911-020-01332-6 (2020).
Ghassemi, M., Oakden-Rayner, L. & Beam, A. L. The false hope of current approaches to explainable artificial intelligence in health care. Lancet Digit. Health. 3 (11), e745–e750. https://doi.org/10.1016/s2589-7500(21)00208-9 (2021).
Mörch, C. M. et al. Artificial intelligence and ethics in dentistry: A scoping review. J. Dent. Res. 100 (13), 1452–1460. https://doi.org/10.1177/00220345211013808 (2021).
World Health Organization. Ethics and Governance of Artificial Intelligence for Health (World Health Organization, 2021).
Xu, Q. et al. Interpretability of clinical decision support systems based on artificial intelligence from technological and medical perspective: a systematic review. J. Healthc. Eng. 2023, 9919269. https://doi.org/10.1155/2023 (2023).
Davis, F. D. Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q. 13 (3), 319–340. https://doi.org/10.2307/249008 (1989).
Venkatesh, V. & Davis, F. D. A theoretical extension of the Technology Acceptance Model: four longitudinal field studies. Manage. Sci. 46 (2), 186–204. https://doi.org/10.1287/mnsc.46.2.186.11926 (2000).
Venkatesh, V. & Bala, H. Technology Acceptance Model 3 and a research agenda on interventions. Decis. Sci. 39 (2), 273–315. https://doi.org/10.1111/j.1540-5915.2008.00192.x (2008).
Holden, R. J. & Karsh, B. T. The Technology acceptance model: Its past and its future in health care. J. Biomed. Inf. 43 (1), 159–172. https://doi.org/10.1016/j.jbi.2009.07.002 (2010).
Rahimi, B., Nadri, H., Lotfnezhad Afshar, H. & Timpka, T. A systematic review of the technology acceptance model in health informatics. Appl. Clin. Inf. 9 (3), 604–634. https://doi.org/10.1055/s-0038-1668091 (2018).
Scherer, R., Siddiq, F. & Tondeur, J. The Technology Acceptance Model (TAM): a meta-analytic structural equation modeling approach to explaining teachers’ adoption of digital technology in education. Comput. Educ. 128, 13–35. https://doi.org/10.1016/j.compedu.2018.09.009 (2019).
Sarhan, S. et al. Perception, usage, and concerns of artificial intelligence applications among postgraduate dental students: cross-sectional study. BMC Med. Educ. 25, 856. https://doi.org/10.1186/s12909-025-07544-6 (2025).
Altındağ, A., Uzun, S., Altındağ, Ö. & Orhan, K. Artificial intelligence in dentistry: knowledge, attitudes, and educational readiness among dental students and dentists. BMC Med. Educ. 26 (1), 972. https://doi.org/10.1186/s12909-026-09264-x (2026).
Venkatesh, V., Thong, J. Y. L. & Xu, X. Consumer acceptance and use of information technology: Extending the unified theory of acceptance and use of technology. MIS Q. 36 (1), 157–178. https://doi.org/10.2307/41410412 (2012).
Venkatesh, V., Morris, M. G., Davis, G. B. & Davis, F. D. User Acceptance of information technology: Toward a unified view. MIS Q. 1 (3), 425–478. https://doi.org/10.2307/30036540 (2003).
Faul, F., Erdfelder, E., Buchner, A. & Lang, A. G. Statistical power analyses using G*Power 3.1: tests for correlation and regression analyses. Behav. Res. Methods. 41 (4), 1149–1160. https://doi.org/10.3758/BRM.41.4.1149 (2009).
Cohen, J. Statistical Power Analysis for the Behavioral Sciences, 2nd edn (Routledge, 1988).
Hair, J. F. Jr, Black, W. C., Babin, B. J. & Anderson, R. E. Multivariate Data Analysis, 8th edn (Cengage Learning EMEA, 2019).
Henseler, J., Ringle, C. M. & Sarstedt, M. A new criterion for assessing discriminant validity in variance-based structural equation modeling. J. Acad. Mark. Sci. 43 (1), 115–135. https://doi.org/10.1007/s11747-014-0403-8 (2015).
Fornell, G. & Larcker, D. F. Evaluating structural equation models with unobservable variables and measurement error. J. Mark. Res. 18 (1), 39–50. https://doi.org/10.2307/3151312 (1981).
Hu, L. T. & Bentler, P. M. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct. Equ. Model. 6 (1), 1–55. https://doi.org/10.1080/10705519909540118 (1999).
Tanriver, G., Soluk Tekkesin, M. & Ergen, O. Automated detection and classification of oral lesions using deep learning to detect oral potentially malignant disorders. Cancers (Basel). 13 (11), 2766. https://doi.org/10.3390/cancers13112766 (2021).
Rajaram Mohan, K. & Mathew Fenn, S. Artificial intelligence and its theranostic applications in dentistry. Cureus 15 (5), e38711. https://doi.org/10.7759/cureus.38711 (2023).
Tetik, G., Türkeli, S., Pinar, S. & Tarim, M. Health information systems with technology acceptance model approach: a systematic review. Int. J. Med. Inf. 190, 105556. https://doi.org/10.1016/j.ijmedinf.2024.105556 (2024).
Topol, E. J. High-performance medicine: the convergence of human and artificial intelligence. Nat. Med. 25 (1), 44–56. https://doi.org/10.1038/s41591-018-0300-7 (2019).
Pashkov, V. M., Harkusha, A. O. & Harkusha, Y. O. Artificial intelligence in medical practice: regulative issues and perspectives. Wiad Lek. 73 (12 Pt 2), 2722–2727. https://doi.org/10.36740/WLek202012204 (2020).
Rajpurkar, P., Chen, E., Banerjee, O. & Topol, E. J. AI in health and medicine. Nat. Med. 28 (1), 31–38. https://doi.org/10.1038/s41591-021-01614-0 (2022).
Joseph, A. M. et al. Factors influencing large language model adoption among dental students: a cross-sectional study. Sci. Rep. 16, 17113. https://doi.org/10.1038/s41598-026-47512-8 (2026).
D´Albis, G. & Capodiferro, S. AI in oral surgery: promise and prudence. Br. Dent. J. 240, 12. https://doi.org/10.1038/s41415-025-9509-2 (2026).
Jacob, C., Sanchez-Vazquez, A. & Ivory, C. Understanding clinicians’ adoption of mobile health tools: a qualitative review of the most used frameworks. JMIR Mhealth Uhealth. 8 (7), e18072. https://doi.org/10.2196/18072 (2020).
Umer, F. Artificial intelligence and emerging technologies in diagnosis of oral potentially malignant disorders. BDJ Team. 11, 454–456. https://doi.org/10.1038/s41407-024-2820-y (2024).
Thorat, V., Rao, P., Joshi, N., Talreja, P. & Shetty, A. R. Role of artificial intelligence (AI) in patient education and communication in dentistry. Cureus 16 (5), e59799. https://doi.org/10.7759/cureus.59799 (2024).
Huang, Y. et al. Mapping artificial intelligence research in oral and maxillofacial surgery: a bibliometric analysis. Int. Dent. J. 76 (3), 109456. https://doi.org/10.1016/j.identj.2026.109456 (2026).
来源:Nature 健康护理主题 · nature.com