Introduction

The rapid evolution of digital dentistry has significantly transformed the accuracy of oral and maxillo­facial lesion diagnosis in recent years, largely driven by imaging advancements, such as cone-beam computed tomography (CBCT) [1]. Over the past decade, diagnostic workflows have shifted from conventional visual assessments to digitally enhanced, AI-driven image interpretation, providing clinicians with unprecedented levels of precision in evaluating oral pathologies. This digital transition has not only refined lesion characteri­zation, but also redefined clinical decision-making in oral healthcare [2].

Within this broader digital era, radiomics has emerged as a pivotal subset of artificial intelligence. Radiomics is a quantitative imaging biomarker science that converts voxel-level data in a radiographic image into high-dimensional features, with potential diagnostic and prognostic value. By extracting and quantifying high-dimensional features from CBCT and other imaging modalities, radiomics transforms raw image data into a wealth of objective biomarkers [3-5]. These features, often imperceptible to the human eye, can be correlated with clinical outcomes [6], which translates the physical and structural properties of dental lesions into actionable digital insights, deepening our understanding of their pathophysiology [7, 8]. While general artificial intelligence (AI)-based image analysis typi­cally focuses on pattern recognition, classification, and auto­mated decision support, radiomics is fundamentally
different in both purpose and methodology. Radio­mics serves as a quantitative imaging biomarker science that systematically uses every information from every picture element of any imaging modality, and converts them into mathematically defined features, enabling objective measurement of tissue heterogeneity, shape, texture, and intensity patterns. Hence, unlike conventional AI methods that rely on learned representations, radiomics emphasizes handcrafted, quantifiable biomarkers. This can be further correlated with biological behavior, clinical outcomes, and prognostic indicators. This distinction positions radiomics not merely as an analytical tool but as a bridge between medical imaging, pathology, and precision medicine, thus transforming from conventional qualitative interpretation of an image to a quantitative one.

Despite its transformative promise within digital dentistry, radiomics remains largely underrecognized and underutilized in routine clinical practice. Limited knowledge, lack of training, and absence of standardized integration frameworks have decelerated its adoption by dental professionals. Against this backdrop, the present study investigated the knowledge, attitudes, and practices (KAP) of oral radiology specialists, gene­ral dental practitioners, and dental students with respect to radiomics. By comparing these distinct professional groups, the study highlighted variations in awareness and readiness to integrate radiomics into digital workflows. As the first investigation of its kind in dentistry, it underlined the urgent need for global standardization of radiomics training, and illustrated its potential to serve as a transformative force in the future of digital dental diagnostics.

Objectives

To assess and compare the knowledge, attitudes, and practices of oral radiology specialists, general dental practitioners, and dental students regarding radiomics in digital dentistry, and to evaluate their readiness for its clinical adoption.

Material and methods

Study design and participants

This was a cross-sectional, questionnaire-based study conducted among dental professionals in North India, including oral radiology specialists, general dental practitioners, and dental students. The study aimed to assess their KAP regarding radiomics in digital dentistry. Participants were recruited using a convenience sampling approach across academic institutions, private dental clinics, and hospitals. To ensure diverse representation, individuals from different professional backgrounds and experience levels were included.

Ethical compliance

The study received ethical approval from the Institutional Review Board of the Teerthanker Mahaveer Dental College and Research Centre. Informed consent was obtained from all participants before data collection. The survey was anonymized to reduce response bias, and all data handling adhered to institutional and international ethical guidelines for participant privacy and confidentiality.

Questionnaire development and validation

A self-structured 23-item questionnaire was developed using Google Forms (Google LLC, Mountain View, CA, USA) and underwent content validation by a panel of five experts in oral radiology and dental public health to ensure clarity, relevance, and scientific rigor. Prior to large-scale distribution, the questionnaire was pilot tested on 30 respondents, consisting of 10 oral radio­logy specialists, 10 general dentists, and 10 dental students, to assess reliability and clarity, with the pilot study demonstrating high internal consistency (Cronbach’s a = 0.85), thereby confirming its suitability. The fina­lized questionnaire comprised four sections: (1) demographic information, including gender, age, professional status, years of experience, institution type, prior AI or machine learning (ML) training, and access to advanced imaging; (2) knowledge assessment, which contained items on familiarity with radiomics, its applications, relevant imaging modalities, and self-assessed knowledge; (3) attitude evaluation, with statements regarding beliefs about the benefits of radiomics, the need for training, and its integration into clinical practice and dental education; and (4) practice and implementation, assessing CBCT usage, utilization of radiomics features, barriers to adoption, and willingness to undergo training. Skip logic was incorporated to maintain a logical response flow, and participants who had never used radiomics (Q20 = No) were directed to identify barriers to adoption (Q21), while those who had used radiomics (Q20 = Yes) bypassed this item and proceeded directly to Q22. To determine an adequate sample size, a power analysis was conducted using G*Power 3.1, which indicated that a minimum of 300 participants was required, based on an assumed effect size (Cohen’s d) of 0.3, an alpha level of 0.05, and a power of 0.95, thereby ensuring sufficient statistical power to detect meaningful differences in KAP scores across professional groups.

Data collection and statistical analysis

The questionnaire was distributed online via email and social media between July and September 2024. Responses were collected anonymously and exported to PASW Statistics for Windows, version 18.0 (SPSS Inc., Chicago, IL, USA) for analysis. Descriptive statistics, including frequencies, percentages, and mean ± standard deviation were used to summarize demographic characteristics and response distributions. Inferential statistical analyses were conducted to examine relationships and differences between professional groups. Pearson’s c² test was applied to assess associations between professional status and categorical responses, while Kruskal-Wallis H test was used to compare KAP scores across groups. Binary logistic regression was performed to identify predictors of high radiomics knowledge, with results reported as odds ratios (OR) and 95% confidence intervals (CIs). For logistic regression, ‘high radiomics knowledge’ was operationalized using a median split of total knowledge score, where participants scoring above the median were categorized as having high knowledge. Hosmer-Lemeshow test was employed to evaluate the goodness-of-fit of the logistic regression model, ensuring its validity. For all statistical tests, p < 0.05 was considered statistically significant.

Results

Descriptive statistics

The study included a total of 300 participants, equally distributed among oral radiology specialists (n = 100, 33.3%), general dental practitioners (n = 100, 33.3%), and dental students (n = 100, 33.3%). The gender distribution was nearly equal, with 153 male respondents (51%) and 147 female participants (49%). The mean professional experience among practitioners, excluding students, was 5.7 ± 2.3 years. Access to advanced imaging technologies varied among groups, with 65% of specialists, 42% of general dentists, and only 18% of students reporting access to CBCT or magnetic resonance imaging (MRI) facilities.

In terms of prior training in AI or ML, 19% of specialists, 11% of general dentists, and only 4% of students had received some form of exposure. Knowledge about radiomics was generally low across all groups, with only 10.3% of all participants reporting being “Familiar” with the concept, while 65.7% stated they were “Not familiar” (Figure 1). On a normalized knowledge scale, oral radio­logy specialists demonstrated higher composite knowledge levels (47.3%) compared to general dental practitioners (6.8%) and dental students (0.4%) (p < 0.001). Additionally, 67% of specialists correctly identified radiomics’ application in disease detection compared to 48% of general dentists and 22% of students (p = 0.002).

Attitude toward radiomics was generally positive, with 76% of all participants agreeing that radiomics could enhance diagnostic accuracy. A significantly higher proportion of specialists (94%) and general dentists (87%) supported integrating radiomics into dental curricula compared to students (72%, p = 0.007). Similarly, willingness to undergo radiomics training differed significantly across groups, with 84% of specialists, 73% of general dentists, and only 61% of students expressing willingness (p = 0.003).

Regarding practical application of radiomics, only 9% of all participants had ever used radiomics features for clinical diagnosis. Among those who had not used radiomics, the most commonly cited barriers were lack of training (61%), followed by limited software access (24%) and lack of awareness (15%). Specialists reported the highest frequency of CBCT usage (83%) compared to 67% of general dentists and only 49% of students (p < 0.001). Furthermore, specialists were significantly more likely to express intent to integrate radiomics-based analysis into their workflow compared to general dentists and students (p = 0.001).

Reliability analysis

The internal consistency of the 23-item questionnaire was evaluated using Cronbach’s a, yielding an overall value of 0.85, indicating high reliability. Each section of the questionnaire demonstrated acceptable reliability, with a = 0.82 for the knowledge section, a = 0.84 for the attitude section, and a = 0.81 for the practice section. All individual items within each subscale exhibited corrected item-total correlations exceeding 0.40, further confirming the internal consistency of the instrument. These results indicated that the questionnaire items effectively measure the intended constructs, and can be reliably used for assessing knowledge, attitudes, and practices related to radiomics in dentistry (Table 1).

Association between professional status and KAP domains

The Pearson’s c² test was used to determine associations between professional status and various categorical responses. The results indicated a statistically significant relationship between professional status and familiarity with radiomics (c² = 25.34, p < 0.001), confirming that knowledge of radiomics varies significantly across diffe­rent professional groups. Specialists were more likely to be familiar with radiomics, whereas students demonstrated the lowest familiarity (Figure 2). Likewise, a significant association was found between prior AI/ML training and familiarity with radiomics (c² = 31.07, p < 0.001), indicating that prior exposure to artificial intelligence plays a crucial role in understanding radiomics applications.

Furthermore, a statistically significant difference was observed in willingness to undergo radiomics training across professional groups (c² = 18.72, p = 0.003), with specialists showing the highest willingness. However, gender did not have a significant impact on the willing­ness to use radiomics software (c² = 3.42, p = 0.18). These findings highlight the importance of professional background and prior training in shaping radiomics knowledge and attitudes (Table 2).

Comparison of KAP scores

The Kruskal-Wallis H test was conducted to compare KAP scores across professional groups. Significant diffe­rences were observed in all three domains, with “Knowledge” scores showing the highest variation (H = 32.51, p < 0.001). Post-hoc pairwise comparisons using Dunn’s test with Bonferroni correction revealed that oral radiology specialists scored significantly higher than both gene­ral dentists and students regarding knowledge.

Moreover, “Attitude” scores demonstrated statistically significant differences (H = 21.86, p = 0.002), with specialists displaying the most positive attitudes toward radiomics education and integration. Similarly, “Practice” scores were significantly higher among specialists (H = 29.72, p < 0.001), largely due to their increased frequency of CBCT use and greater familiarity with radiomics applications. These results reinforce the notion that professional background influences radiomics-related KAP implementation (Table 3).

Predictors of high radiomics knowledge

To identify factors predicting high radiomics knowledge, a binary logistic regression model was used. Prior AI/ML training was the strongest predictor of high radiomics knowledge (OR = 3.85, 95% CI: 2.20-6.70, p < 0.001). Access to advanced imaging was also independently associated with high radiomics knowledge (OR = 2.55, 95% CI: 1.45-4.50, p = 0.001).

Participants with 5 years or more of professional experience had higher odds of high radiomics knowledge (OR = 2.10, 95% CI: 1.30-3.45, p = 0.002). Affiliation with an academic institution was likewise a significant predictor (OR = 1.80, 95% CI: 1.05-3.10, p = 0.035). Thus, prior AI/ML training, access to advanced imaging, greater professional experience, and academic institutional affiliation were significant predictors of high radiomics knowledge (Table 4).

Validation of the logistic regression model

The goodness-of-fit of the logistic regression model was assessed using the Hosmer-Lemeshow test, which yielded a c² value of 6.45 (p = 0.59). The non-significant p-value indicated a good model fit, suggesting that the predicted probabilities align well with the observed data. This confirms the robustness of the regression model in predicting factors influencing radiomics knowledge.

Discussion

This study provides crucial insights into the KAP regarding radiomics among dental professionals in northern India, including oral radiology specialists, general practitioners, and dental students. As the first investigation of its kind, it not only captures the present state of awareness, but also delineates the barriers and enablers influencing the clinical integration of radiomics. Viewed from a digital dentistry perspective, these findings highlight both the untapped potential of radiomics and the urgent need to embed it within the broader digital transformation of oral healthcare.

A key finding of this study is the significant knowledge gap, with 65.7% of participants reporting unfamiliarity with radiomics. This deficit was especially marked among general practitioners and dental students, suggesting limited exposure to radiomics during dental training. The absence of formal teaching in this domain echoes trends reported in the United States and Europe, where professionals expressed similar concerns about inadequate training in radiomics and other AI-driven innovations [9, 10]. A global review by Leite et al. [15] further emphasized the widespread deficiency of AI content in dental curricula. This international consensus indicates that unless dental education frameworks are modernized to prioritize digital competencies, professionals will remain underprepared for the realities of digitally augmented diagnostics. The findings therefore advocate for the urgent integration of radio­mics into structured training pathways, not as an elective addition but as a core component of digital dentistry.

Interestingly, despite this knowledge gap, participants demonstrated overwhelmingly positive attitudes toward radiomics integration, with 98% expressing willingness to adopt it in dental practice. This enthusiasm aligns with global trends in radiology, where the adoption of AI and radiomics has been strongly supported due to their ability to improve diagnostic precision and efficiency [16, 17, 21]. Thurzo et al. [21] similarly noted that professionals exposed to high-tech diagnostic systems were more receptive to AI adoption. Comparable observations by Pringle et al. [13] and Schwendicke et al. [11] confirm that dental professionals worldwide increasingly recognize the transformative role of AI in clinical care. However, enthusiasm alone cannot translate into practice unless underpinned by accessible training opportunities and standardized clinical guidelines for digital tools, such as radiomics. When responses were compared across age and experience levels, younger participants and those with fewer years of clinical practice demonstrated lower awareness of radiomics, but showed higher willingness to adopt AI-based tools compared to older and more experienced clinicians. This suggests a generational shift wherein digitally exposed trainees may adopt radiomics more rapidly once formal training becomes available, whereas senior practitioners may rely more heavily on prior clinical experience.

The practical challenges identified in this study, namely the lack of training, limited access to radiomic software, and complexity in data interpretation, mirror the barriers faced globally [12]. For instance, Müller et al. [12] documented similar concerns in Germany, particularly regarding high software costs and lack of formal training. These findings reveal that the integration of radiomics into dental practice is not hindered by skepticism but by resource constraints. Overcoming these barriers requires multipronged strategies, such as the development of scalable online certification programs, partnerships with technology companies to design affordable, user-friendly radiomics platforms, and the creation of accessible cloud-based imaging solutions. Such approaches would democratize access to advanced diagnostic tools, aligning with the ethos of digital dentistry as an equalizer, bringing cutting-edge diagnostics into everyday practice.

From a digital dentistry standpoint, the value of radiomics lies in its ability to enrich routine workflows that are already reliant on imaging. By embedding radiomics into the digital workflow of CBCT, intraoral scanning, and CAD/CAM systems, routine diagnostics may transition from subjective interpretation to data-driven precision [19]. This shift does not just improve accuracy but also reduces inter-operator variability, a persistent limitation in conventional diagnostics. Moreover, radiomics can play a transformative role in personalized treatment planning, a cornerstone of digital dentistry. Beyond detecting lesions, radiomics can provide prognostic insights, such as predicting aggressiveness of cystic lesions or tumor recurrence risk, which can guide surgical planning and follow-up protocols. In orthodontics and implantology, radiomics-derived metrics could be used to simulate treatment outcomes with greater reliability, strengthening the predictive power of digital models. These applications illustrate that radiomics is not an isolated innovation, but part of a continuum of digital dentistry tools that collectively move the field toward precision, efficiency, and patient-centered care [20]. Within the era of digital adoption in dentistry, these findings also reflect an early stage of radiomic readiness, where clinicians are comfortable with foundational digital tools but have limited exposure to quantitative imaging methods. As digital imaging, CAD/CAM systems, and AI-assisted segmentation become routine, radiomics naturally emerges as the next layer of capability, extending these established technologies with voxel-level quantitative insights that support more precise and personalized diagnostic decisions.

The translational evidence from oncological field further reinforces this perspective. Liu et al. [22] demonstrated that radiomics can convert routine medical images into high-dimensional quantitative data, which enhance diagnostic precision and enable personalized prediction models in clinical practice, especially while dealing with head and neck cancer. Applying these principles to dentistry suggests similar potential for improving lesion characterization, reducing subjective variability, and strengthening data-driven chairside decision-making. Likewise, Limkin et al. [23] emphasized that the successful implementation of radiomics requires standardized imaging workflows, reproducible feature extraction, robust data management, and effective multidisciplinary collaboration. These considerations directly parallel the challenges identified in the current study, highlighting the need for structured training, consistent imaging protocols, and accessible radiomics platforms to support meaningful integration into dental practice.

The enthusiasm for further research expressed by participants reflects a collective recognition of radiomics’ transformative potential. Semerci et al. [14] point out the importance of making AI technologies intuitive and accessible for dental professionals, while Musleh et al. [18] stressed the need for standardized data extrac­tion protocols to facilitate seamless integration into workflows. These perspectives align with the present findings, featuring that radiomics will only become clinically meaningful when paired with structured training, supportive software ecosystems, and practical workflow integration.

However, the main limitation of this study is its reliance on self-reported responses from dental professio­nals in northern India, which may not accurately reflect actual clinical adoption and restricts the generalizability of results to other regions. To address this, future research should employ observational or interventional designs across diverse practice settings to more accurately capture readiness and real-world adoption of radiomics in dentistry. Another limitation is that the present study focused primarily on CBCT-based radiomics. This limits the generalizability of radiomics’ applicability across other imaging modalities; future investigations incorporating MRI, CT, intraoral scanners, and emerging optical modalities will provide a more comprehensive understanding of radiomics integration in dentistry.

Conclusions

Radiomics represents a pivotal advancement within the broader landscape of digital dentistry, offering the potential to move routine diagnostics beyond subjective interpretation toward data-driven precision. By quantifying imaging features imperceptible to the human eye, radiomics can enhance early detection of oral pathologies, refine prognostic assessment, and support persona­lized treatment planning across specialties, such as oral radiology, implantology, orthodontics, and dental onco­logy. Although current adoption is limited by knowledge gaps, software accessibility, and integration challenges, the strong willingness of dental professionals to embrace this technology highlights a clear path forward. With targeted training, user-friendly platforms, and standardized clinical guidelines, radiomics can become an integral component of digital workflows, ultimately transforming everyday dental practice into a more precise, predictive, and patient-centered discipline.

Disclosures

Funding: This research received no external funding.

Institutional Review Board statement: The study was approved by the Ethics Committee of the Teerthanker Mahaveer Dental College and Research Centre (approval No. TMDCRC/IEC/PHD/2425/DENTAL02; IEC Proposal No. S002/24), dated 04/04/2025. All participants provided informed consent prior to participation, ensuring voluntary involvement and confidentiality of responses. The study adhered to the ethical standards outlined in the Declaration of Helsinki.

Informed consent statement: Informed consent was obtained from all participants prior to participation. All participants provided consent for the use of this nonidentifiable information in aggregate form for publication purposes.

Data availability statement: All the data generated or analyzed in this study are included in this manuscript. The data generated in this study may be requested from the author.

Acknowledgments: None.

Conflicts of interest: The author declares no conflicts of interest.

AI use statement: Not applicable.

References