Attitude of Brazilian dentists and dental students regarding the future role of artificial intelligence in oral radiology: a multicenter survey

Abstract

Objectives:

The aim of this study was to assess the attitude of dentists and dental students in Brazil regarding the impact of artificial intelligence (AI) in oral radiology, and to evaluate the effect of an introductory AI lecture on their attitude.

Methods:

A questionnaire was prepared, comprising statements regarding the future role of AI in oral radiology and dentistry. A lecture of approx. 1 h was prepared, comprising the basic principles of AI and a non-exhaustive overview of AI research in medicine and dentistry. Participants filled in the questionnaire prior to the lecture. After the lecture, the questionnaire was repeated.

Results:

Throughout 7 sessions at 6 locations, 293 questionnaires were collected. The majority of participants were undergraduate dental students (57%). Prior to the lecture, there was a strong agreement regarding the various future roles and expected impact of AI in oral radiology. Approximately, one-third of participants was concerned about AI. After the lecture, agreement regarding the different roles of AI in oral radiology increased, overall excitement regarding AI increased, and concerns regarding the potential replacement of oral radiologists decreased.

Conclusions:

A generally positive attitude towards AI was found; an introductory lecture was beneficial towards this attitude and alleviated concerns regarding the effect of AI on the oral radiology profession. Given the unprecedented, ongoing revolution of AI-augmented radiology, it is pivotal to incorporate AI topics in dental training curricula.

Introduction

In the last few years, artificial intelligence (AI) has become the hottest topic in radiology.¹ New AI methods using deep learning (DL), often involving convolutional neural networks (CNNs),² along with improved computational capacity has resulted in the exploration of various AI applications. The most commonly investigated applications involve the detection of pathosis and the segmentation of anatomical and pathological structures.³

In dentistry, AI applications have been investigated as early as 1992.^3,4 Whereas most of the research in subsequent years involved cephalometric landmark detection, CNN-based DL has recently been investigated for tasks such as osteoporosis detection, tooth detection/numbering, and the detection of various types of oral pathosis.^2,3,5,6 Seeing that most of the AI research in radiology involve imaging techniques with large patient populations, such as chest radiography and mammography, dental radiology is highly suited for AI because of the high frequency of imaging performed worldwide. Based on data of the United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR),⁷ the European Commission’s Radiation Protection report 180 [2] and the 2014–2015 NEXT survey in the United States [3],^7–9 a conservative estimate has been made that at least 1.5 billion dental radiographic examinations are performed annually at the time of writing.¹⁰

In the context of this ongoing AI revolution, it is essential to ensure that current and future clinical practitioners are up-to-date regarding the current state and potential of this technology. Whereas the number of published research studies on AI in radiology is increasing considerably, evidence-based guidelines and position statements are pending, especially in dentistry. A lack of information, or an excess of misinformation, regarding the role of AI in the future of clinical practice could lead to negative attitudes and poorly informed career choices.¹¹ On the other hand, the limited sample sizes in the majority of currently published AI studies in dentistry indicate that, for many applications, clinical implementation is several steps away.^2,3 The overall consensus, however, seems to be that it is a matter of time before many radiological tasks are “augmented”¹² using AI, with the extent of the AI’s role being highly task-specific. Thus, it is important to provide students and clinicians with accurate, objective and up-to-date information during this transitional phase. In this context, it is valuable to assess the current attitude of clinical dentists as well as undergraduate students in dentistry regarding the future role of AI. Therefore, the aim of this study was to assess the attitude of (future) dentists regarding the use of AI in oral radiology, as well as the effect of an introductory AI lecture on their attitude.

Methods and materials

A questionnaire was prepared based on the study from Pinto dos Santos et al¹³. The survey included demographic information, professional background, familiarity with AI, applicability of AI in oral radiology/dentistry, and attitude regarding the impact of AI in oral radiology/dentistry. The questionnaire was provided to participants of seminars or congresses at six different locations in Brazil: (1) University of São Paulo, School of Dentistry of Ribeirão Preto (Ribeirão Preto, SP); (2) University of São Paulo, School of Dentistry of Bauru (Bauru, SP); (3) Federal University of Espírito Santo (Vitória, ES); (4) XXIth annual meeting of SNNPqO (Campina Grande, PB); (5) State University of Northern Paraná (Jacarezinho, PR); (6) Faculty of Medicine and Dentistry São Leopoldo Mandic (Campinas, SP). The participants comprised a heterogeneous group of students, clinicians, researchers and professors from local institutions. Participation in the survey was voluntary, anonymous and was not part of any student evaluation. Hence, no ethical approval was required.

As the aforementioned study by Pinto dos Santos et al¹³ was aimed at undergraduate medical students, the following adjustments were made to make it more suitable for the field of oral radiology:

• Answer options regarding professional background were adapted to represent different types of students and professionals in dentistry.

• A question was added to ask licensed dentists regarding their specialty, if applicable.

• Questions pertaining ‘radiology’ were specified to ‘oral radiology’, and ‘medicine/physician’ was replaced with ‘dentistry/dentist’.

• Answer options regarding the applications of AI in oral radiology were expanded by adding ‘treatment planning’ and ‘image processing’. Specific examples were added for most of the options to ensure that the difference between them was clear.

• Specific cut-off dates were provided in questions regarding the potential replacement of oral radiologists (≤15 years from now) and dentists in general (≤30 years from now) by AI-based systems.

The original questionnaire was prepared in English; it was translated into Portuguese by a native speaker.

A lecture of approx. 1 h was prepared, comprising the following topics:

• Introduction: examples of AI in fiction and current society (5’).

• Overview of the basic principles of AI: examples of animal recognition, showing improved AI performance over time and indicating the limitations of conventional image processing approaches for complex recognition tasks (5’).

• Basic concepts: distinction between AI/machine learning/DL, classification vs regression tasks (with radiological example), different types of AI-based learning (supervised/unsupervised/reinforcement) (10’).

• Supervised learning: example of a diagnostic application, data labelling, concepts related to training (fitting a model with weights, loss, gradient descent, moment, data augmentation), brief illustration of variety of machine learning algorithms (10’).

• Convolutional neural networks: basic concepts, benefits for image-based tasks (10’).

• Non-exhaustive overview of published research studies on AI applications in medicine and dentistry. The selected studies comprised various types of algorithms, and types of tasks, e.g. diagnosis, segmentation, image enhancement, prediction of patient outcome (15’).

• Future prospects: steps towards AI implementation in radiology, how to get involved in AI research (5’).

Care was taken to provide all information in an objective manner, as to not to steer the audience towards any particular opinion regarding the role of AI. The lecture was given in English by a single presenter throughout all sessions. A PDF file of the lecture can be provided by the corresponding author at request. After the presentation, questions and answers were mostly in English, although a translator was available. Immediately after the lecture, the participants were asked to fill in the last part of the questionnaire, which repeated the questions pertaining their view on the applications and role of AI in dentistry. Results were analyzed using descriptive statistics. Answers prior to and subsequent to the lecture were compared using the Wilcoxon matched-pairs signed rank test. Furthermore, the answers regarding the attitude towards AI were split up using subpopulations based on gender, professional level, dental specialty, and self-reported technological proficiency. To assess differences in response due to professional level, undergraduate students were compared with all other participants. To assess the effect of the dental specialty, a three-way intercomparison was performed between general practitioners, specialists in oral radiology, and other dental specialists. In terms of technological proficiency, participants who answered “agree” or “rather agree” on this question were compared with others. The Mann–Whitney test was used for all comparisons between subpopulations. A significance level of 0.05 was used for all tests.

Results

Throughout 7 sessions at the aforementioned 6 locations, 293 questionnaires were collected. General data regarding the participants are summarized in Table 1. The majority of the participants were undergraduate students in dentistry (57.0%), followed by postgraduate/PhD students (20.2%) and professors (14.7%). Out of the participants with a degree in dentistry, the majority were specialists in fields other than oral radiology (52.3%) (Table 2). The age of the participants was 22.6 ± 4.1 years for undergraduate students and 34.3 ± 11.9 years for others; 69% of the participants were female (Table 3). A symmetric distribution was found regarding the participants’ self-assessment of their technological proficiency, with a small number finding themselves decidedly proficient (9.4%) or inept (7.7%).

Table 1.

Professional background of participants

Background	Participants	%
Undergraduate student (dentistry)	167	57.0
Undergraduate student (other)	5a	1.7
Postgraduate/PhD student (oral radiology)	30	10.2
Postgraduate/PhD student (other dental disciplines)	29	9.9
Postdoctoral researcher	5	1.7
Professor	43	14.7
Dentist (clinical practitioner & none of the above)	12	4.1
Other	2b	0.7
Total	293

^aMedicine (n = 2), biology (n = 1), biotechnology (n = 1), computer science (n = 1).

^bSystems analyst (n = 1), unspecified researcher (n = 1).

Table 2.

Distribution of participants with dental license

	Participants	%
General practitioner	21	19.3
Specialist (oral radiology)	30	27.5
Specialist (other)	57	52.3
Prosthetic technician	1	0.9
Total	109

Table 3.

Demographics and self-assessment of technological proficiency

	Median	First/third quartile	Min/Max
Age (years)	24	21/32	18/63
	Male	Female	N/A
Gender (%)	89 (30.4%)	199 (67.9%)	5 (1.7%)
	Agree entirely	Rather agree	Rather disagree	Disagree entirely
Technologically proficient (%)	9.4	43.2	39.7	7.7

The participants’ overall awareness of AI is shown in Table 4. Whereas 63.0% of the participants was unfamiliar with the application of AI in radiology, a substantial amount (24.7%) assessed that they already have a basic understanding of this technology. Most of the prior awareness of AI came from media (93.3%), including social media (90.0%). Almost half of the participants had already attended some type of presentation regarding AI (not counting the lecture they attended on the day of the survey).

Table 4.

Prior awareness of artificial intelligence and deep learning

“Deep Learning” and “Artificial Intelligence” are currently being broadly discussed in the radiological community	Yes
Were you already aware of these topics in radiology?	37.0%
Do you personally have a basic understanding of the technologies used in these topics?	24.7%
Other applications we use in daily life already use artificial intelligence (e.g. speech-/text-recognition, spam-filters, recommendation algorithms). Were you aware of this?	Yes
From the media	93.3%
From social media	90.0%
From lectures	48.5%
From friends/family	46.8%

Attitude regarding AI and effect of introductory lecture

Opinions regarding the potential applications of AI in oral radiology are found in Table 5. Percentages in the following subsection always refer to absolute, not relative, changes in response frequency (e.g. a change in response frequency from 40 to 60% would be indicated as ’+20%’, not ‘+50%’). Prior to the lecture, most of the participants partially or fully agreed that AI could play a role in automatic detection of pathosis on images, treatment planning, image processing and selection of appropriate imaging techniques. However, only a slight majority (62.6%) agreed that it could provide the final diagnosis. After the lecture, a significantly increased agreement was found for each potential AI application except treatment planning.

Table 5.

Answers regarding applications of AI in oral radiology

What potential applications for AI in oral radiology do you see?
	Before/after lecture	Agree entirely (%)	Rather agree (%)	Rather disagree (%)	Disagree entirely (%)	p-value
Automated detection of pathologies in imaging exams	Before	50.3	40.9	7.7	1.0
	After	62.5	33.3	3.8	0.4
	Difference	+12.1	−7.6	−3.9	−0.7	<0.001
Automated final diagnosis from imaging exams	Before	27.6	35.0	31.5	5.9
	After	34.9	34.1	22.6	8.4
	Difference	+7.2	−0.9	−8.9	+2.5	<0.05
Treatment planning (e.g. selection and positioning of dental implant, risk evaluation for third molar extraction)	Before	50.9	39.7	8.4	1.0
	After	50.0	38.1	10.0	1.9
	Difference	−0.9	−1.6	+1.6	+0.9	0.69
Image processing (e.g. cephalometric tracing, tooth numbering, delineation of anatomical structures)	Before	64.7	30.1	5.2	0.0
	After	73.6	24.1	1.9	0.4
	Difference	+8.9	−5.9	−3.3	+0.4	<0.01
Automated indication of appropriate imaging exams	Before	37.3	44.0	18.3	0.4
	After	52.0	37.1	9.4	1.6
	Difference	+14.6	−6.9	−8.9	+1.2	<0.0001

p-values refer to paired evaluation of answers before and after an introductory lecture on AI (Wilcoxon matched-pairs signed rank test).

Responses regarding the participants’ personal perception regarding the use of AI in oral radiology and dentistry are found in Table 6. Before the lecture, for most of the questions, a high consensus (>85% agreement or disagreement) was seen for most statements; mixed responses were found regarding being worried by AI development (33.5% agreed) and the substitution of oral radiologists by AI programs in the next 15 years (22.9% agreed). The most notable effects of the lecture were:

• Over 11% of the responders switched from being somewhat agreed to completely agreed that AI represents a revolution in oral radiology, and 5.5% of binary responses switched to agreeing that it will revolutionize dentistry (p < 0.01).

• Interestingly, there was a non-significant tendency to switch from slightly disagree (−16.2%) to completely disagree (+14.2%) for the question of whether oral radiologists will be replaced by AI. On the other hand, agreement regarding long-term (<30 years) replacement of dentists by AI-based alternatives increased by 8.2% (p < 0.001), and agreements that dentists will never be replaced decreased by 7.4% (p < 0.01).

• There was a generally increased excitement regarding the use of AI in oral radiology (+5.6%; p < 0.001) and dentistry (+5.5%; p < 0.001). On the other hand, there was no significant change in responses regarding whether AI will improve oral radiology or dentistry, although a higher degree of agreement was found for both questions.

• Although there was a high overall agreement that AI should be part of dental training curricula prior to the lecture (94.4%), the percentage of participants with strong agreement significantly increased after the lecture (+7.2%; p < 0.05).

Table 6.

Attitude regarding impact of AI in oral radiology

In your personal opinion, how accurate are the following statements?
	Before/after lecture	Agree entirely (%)	Rather agree (%)	Rather disagree (%)	Disagree entirely (%)	p-value
AI will revolutionize oral radiology	Before	52.1	45.1	2.4	0.3
	After	63.4	33.6	1.9	1.1
	Difference	+11.3	−11.6	−0.5	+0.8	<0.01
AI will revolutionize dentistry in general	Before	48.8	40.8	10.1	0.3
	After	56.7	38.3	4.2	0.8
	Difference	+7.9	−2.5	−5.9	+0.4	<0.01
The human oral radiologist will be replaced within 15 years	Before	4.9	18.0	55.6	21.5
	After	8.4	16.5	39.5	35.6
	Difference	+3.5	−1.5	−16.2	+14.2	0.13
All dentists will be replaced within 30 years	Before	2.5	2.8	30.9	63.9
	After	5.0	8.5	26.9	59.6
	Difference	+2.5	+5.7	−4.0	−4.2	<0.001
These developments worry me	Before	12.1	21.4	31.3	35.2
	After	15.3	22.7	26.7	35.3
	Difference	+3.2	+1.4	−4.7	+0.1	0.41
These developments make oral radiology more exciting to me	Before	33.8	51.4	12.0	2.8
	After	47.3	43.5	6.9	2.3
	Difference	+13.5	−7.9	−5.0	−0.5	<0.0001
These developments make dentistry in general more exciting to me	Before	36.7	50.9	10.7	1.8
	After	46.7	46.3	5.0	1.9
	Difference	+10.1	−4.6	−5.7	+0.2	<0.001
AI will never make the human dentist expendable	Before	70.0	19.8	6.4	3.9
	After	62.5	19.9	9.6	8.0
	Difference	−7.5	+0.1	+3.2	+4.2	<0.01
AI will improve oral radiology	Before	65.2	33.8	0.7	0.3
	After	70.4	27.7	1.5	0.4
	Difference	+5.2	−6.1	+0.8	+0.0	0.19
AI will improve dentistry in general	Before	62.9	35.0	2.1	0.0
	After	69.0	29.1	1.5	0.4
	Difference	+6.0	−5.8	−0.6	+0.4	0.05
AI should be part of dental training	Before	54.9	39.6	5.2	0.3
	After	62.1	33.0	4.2	0.8
	Difference	+7.2	−6.6	−1.0	+0.4	<0.05

AI, artificial intelligence.

p-values refer to paired evaluation of answers before and after an introductory lecture on AI (Wilcoxon matched-pairs signed rank test).

Effect of gender, professional level and technological proficiency on attitude regarding AI

The following significant differences were found:

• No significant difference in response between genders, except for the statement that AI will revolutionize dentistry in general, for which females showed higher agreement (p = 0.02).

• Undergraduate students showed a significantly different response regarding the replacement of oral radiologists (higher agreement; p = 0.03), excitement regarding AI’s impact on the dental profession (lower agreement; p = 0.04) and the statement that AI should be part of dental training (lower agreement; p < 0.01).

• General practitioners showed an increased excitement regarding the impact of AI on the dental profession than specialists in oral radiology (p = 0.03) as well as other dental specialists (p < 0.01). Note that the sample size was somewhat limited for these subgroups, which limits the statistical power.

• Participants with technological proficiency answered differently for statements regarding the replacement of oral radiologists (higher agreement, p = 0.01) and excitement regarding AI’s impact on the dental profession (higher agreement, p = 0.02).

Discussion

In the midst of the ongoing AI revolution that is expected to affect profoundly the future of radiology, this study assessed the current perception of AI in oral radiology and dentistry. Whereas there was a generally optimistic attitude towards AI, it was found that an introductory lecture on this topic had an immediate, positive effect on the perception of dental students and professionals.

A degree of concern and uncertainty by (present and future) dental professionals can be expected, considering the growing evidence regarding the performance of AI tools for dedicated tasks.^2,3,5 Furthermore, people have been made aware of the benefits of AI outside of medicine for relatively complex tasks (e.g. autonomous vehicles). Although it is possible to predict the extent of the impact of AI in dentistry, as well as the timeline of this impact, it is conceivable that the role of responsibilities of radiologists will evolve alongside the development of diagnostic AI tools.

Although this is the first survey of this kind in dentistry, several recent studies have assessed attitudes regarding the impact of AI in medical (i.e. non-dental) radiology. The current survey was adapted from the study by Pinto dos Santos et al¹³ involving German medical students. They found that students are aware, but not concerned about the role of AI in radiology or general medicine. A majority of students judged that AI topics should be included in the medical training curriculum. Other studies involving Canadian¹⁴ and Swiss¹⁵ medical students showed a higher degree of concern, indicated by their reduced intention of choosing radiology as a specialty due to their uncertainty regarding the future, and possible obsolescence, of this profession within a few decades. Similarly, two separate surveys involving US- and UK-based medical students showed a conviction in the future role of AI, coupled with a decreased interest in radiology by almost half of the students.^16,17 In the US study, it was found that most of the current awareness of AI comes from online articles, which can be expected to result in a more negative outlook due to overhyping.¹⁶ In the UK study,¹⁷ participants that received prior teaching in AI were less adverse towards choosing radiology as a profession, indicating the positive impact that can be achieved by providing objective and realistic information regarding AI.¹¹

Another US survey at a radiology residency program revealed a lack of acquaintance to the state of the art in medical AI, leading to concerns by trainees regarding the impact of AI.¹⁸ A French survey comprising radiologist residents and senior radiologists showed similar sentiments regarding a lack of information in AI, and a willingness to familiarize themselves with this topic at a theoretical and practical level.¹⁹ The French participants were generally optimistic regarding the future impact of AI, expecting a reduced error rate and a change in workflow that allows for more patient interaction.¹⁹

Recently, a consortium of European and North American societies released a statement regarding the ethical use of AI in radiology; among others, the need to develop codes of ethics and practice was stressed.²⁰ Several other national associations or communities have released statements that also emphasized the need for teaching^21,22; the French radiology community in particular went into detail regarding the rapid changes that are needed in curricula for radiology residents as well as continuing education.²¹ It is expected (and recommended) that AI tools in dentistry as well as ethical standards, regulations and guidelines, will be developed in parallel to those in medicine.⁵

The adaptation of teaching curricula and guidelines in dentistry, as well as the actual implementation of AI tools in clinical practice, may require a different approach in countries with or without an oral radiology specialty. The current study was performed in a country that has a recognized oral radiology specialty. At the time of writing, it is estimated that 53 countries worldwide recognize this specialty; notably, this is only the case in 3 out of 27 EU member states and 7 of 53 countries in the WHO European Region (private communication, Axel Ruprecht, University of Florida, 2019). In countries with an oral radiology specialty, the ‘augmented radiology’¹² concept will need to be discussed in considerable detail, to ensure that current and future specialists in this field are prepared to make use of AI to its full extent. On the other hand, one could stratify the approach towards teaching, regulations and ethical guidelines among all countries, regardless of the existence of an oral radiology specialty. While it can be envisaged that certain radiological tasks will be gradually performed by AI for the purpose of reduced reporting time and/or higher diagnostic efficacy, this does not imply that such tasks should be removed entirely from radiological courses, regardless of the specialty. The proportion of the curriculum that should be dedicated to AI will undoubtedly be debated in extensive detail in the next years. This is likely to depend on the specialty in question; as a hypothetical example, orthodontists and endodontists in training may require practical training in AI if these tools become part of routine practice in their specialty. Furthermore, undergraduate teaching curricula can contain AI topics regardless of postgraduate career options, although there may be a need for a higher emphasis on the possibly changed, not reduced, role of oral radiologists in the near future to ensure that newly graduated dentists can make an educated choice of their specialty.

Further effort should be undertaken to get a cross-section of the current perception regarding AI in dentistry in different countries and regions. Whereas the sample size in this study was limited due to the requirement of being physically present for a lecture, this was purposely done to assess the immediate effect of such a lecture on the participants’ perception. The use of electronic surveys could be considered as a more wide-scale approach to evaluate the attitude of students and clinicians. Furthermore, the effect of the lecture found in the current study could be reproduced using webinars, if physical lectures are not feasible.