Shifting Risks in Dental Radiography Using Artificial Intelligence

by Peter C. Fritz, B.Sc., D.D.S., F.R.C.D.(C), Ph.D (Perio), M.B.A., LL.M.

Dental Radiography

We are living in a new chapter of human development. Known as the Fourth Industrial Revolution, it represents a fundamental change in how we live, work and relate to one another.1 This change is enabled by extraordinary technology, which blurs the physical, digital and biological worlds in ways that create both prodigious progress and potential peril. The volatility and complexity of this revolution pressure us to rethink how industry sectors and countries will develop as technologies such as gene editing, advanced robotics, and artificial intelligence (AI) converge before our eyes.

Advances in AI and their application to dentistry are increasingly prominent, specifically in dental radiography.2-4 Given the volume of radiographs taken and the importance of radiography on dental diagnosis, dentistry is well-positioned to use AI’s abilities. There are likely far more dental radiographs than any other kind, and once annotated by human experts, they are being used to program AI systems to recognize potential pathology.5 However, as AI becomes responsive to the environment, the intended purpose may evolve into unintended consequences for AI.

If an AI system is not used as intended, what are the potential dangers and risks, and where are they redistributed? This paper will explore how AI-powered dental radiography can have unintended uses, why appropriate safeguards and regulatory frameworks are required to mitigate harms, and which stakeholders should be particularly concerned.


Dental radiography is integral to establishing a dental diagnosis and the primary method for detecting and evaluating decay, periodontal bone loss and other types of dental disease. A problem is a remarkably large disparity among dentists in evaluating radiographs.6 Valchovic et al. measured examiner reliability in non-digital dental radiography and reported a Kappa index of 0.68 to 0.80 for caries and 0.72 to 0.83 for periodontal disease.7 This suggests only a moderate level of agreement among clinicians, with less than 63% of the data considered reliable. This is contrasted by Nguyen et al., who reported that using AI, carious lesions were detected more accurately 75.5 to 93.3%, with a sensitivity of 74.5–97.1%.8 This is a considerable improvement over diagnosis by clinicians using radiographs alone, with “human” sensitivity varying from 19% to 94%.9
AI-powered dental radiography is intended to improve diagnostic efficiency and accuracy significantly. This can help dentists make more accurate diagnoses and develop more effective treatment plans. However, there are several potential dangers or risks associated with using AI in radiography that impacts clinicians.


One potential harm of using AI in radiography is the possibility of errors or biases in the algorithms or data used to train the AI systems. If the dentist automatically defers to the guidance of the AI system, this could lead to incorrect diagnoses and treatment recommendations and thus have serious consequences for patients. This could include unnecessary or harmful treatments or missed opportunities for timely and effective care.

Using diverse and representative datasets to train and validate the algorithms will help reduce bias from AI systems. For example, this can help ensure that AI systems accurately recognize and classify abnormalities in dental images, regardless of patient characteristics. However, using diverse datasets alone may not be sufficient to remove bias from AI systems. Bias can also be introduced through the design and implementation of the algorithms.

Any critiques of bias creeping into AI must recognize that humans make vital decisions that are not always based solely on reason and logic. Dentists can have a range of biases: overconfidence, rule-based thinking, confirmation bias, and a desire to find simple conclusions that explain everything. These humanistic interactions are challenging to digitize. Dentistry is an art and a science, and clinical decisions are made using the head, heart and hand. This is fundamentally why dentists are irreplaceable by machines. Algorithms are cold and inhumane tools; however, when used with experienced clinical judgement and empathy, they can significantly amplify clinician performance.


AI-powered radiography is less accurate than human experts in specific tasks, particularly when dealing with rare or complex cases. In addition, dentists, clinics and dental service organizations will be accountable for decisions made by AI-powered radiography. A human mistake can affect one patient; an AI system error could affect hundreds or thousands. Therefore, dentists must continue to challenge the recommendations of AI-powered algorithms as they are ultimately responsible for the accuracy of the diagnosis and appropriateness of the treatment.


Another potential danger of using AI in dental radiography is the potential for AI to disrupt the traditional roles and responsibilities of dentists. If AI is used to automate decision-making processes, it may change the nature of dentists’ work and the skills and knowledge required to perform it. This could have implications for the training and education of dentists and their professional development. To mitigate this risk, it is crucial for dental regulatory bodies to establish clear guidelines for the use of AI and to provide training and education programs to help dentists adapt to AI use in their practice.



The impact of AI on dental radiography is so profound that educators and regulators must rethink its fundamental competency framework.10 Specifically, the framework should integrate as an essential competency, “digital health literacy,” which is the ability to seek, find, understand, and appraise health information from electronic sources and apply the knowledge gained to addressing or solving a health problem.11

AI-powered dental radiography will become fundamental to the practice of dentistry. To meet patient needs, dentists will need a basic understanding of the technology and the ability to assess suggested results critically. New technology-oriented roles and specialties will emerge, and there is potential to redefine the specialty of dental radiology into one focusing more on clinical informatics.

Educators need to develop guidelines and principles for integrating the teaching of AI and emerging digital technologies across all dental programs and incorporate these teachings as a component of continuous professional development.10



Intended or unintended, AI-powered dental radiography will benefit insurers of dental plans. By using AI to analyze x-ray images, insurers will have more accurate information about the health of their policyholders’ teeth and gums. This could help insurers make more informed decisions about coverage and pricing.

Given the enormous amount of data created daily through dental radiography, it would be relatively easy to study the longevity of specific treatment modalities (i.e., bridge versus an implant). This should influence insurance premiums and policies regarding which clinical treatments to include in the insurance coverage for the best return on investment.

When a treatment plan is presented to a patient, a cumbersome step is to send the treatment plan to the insurance provider to obtain a pre-authorization for the recommended services. Imagine, with the patients’ consent, each radiograph that AI processed was automatically sent to the insurer to allow for immediate approval or denial using a smart contract or similar mechanism. This could create a digital record of the patient’s oral health, potentially stored in a blockchain, verify pre-existing treatment and evaluate the necessity of the planned treatment.12


An unintended consequence of using AI-powered radiography is that insurers could help dentists and their clients analyze dental radiographs to identify and confirm dental issues earlier, potentially conflicting with the dentist’s clinical judgement. This second opinion model could potentially shift the risk of the responsibility of patient care away from the dentist and onto the insurer.

If every dental radiograph were requested by the insurer as a condition of benefits coverage, this would create an enormous amount of patient data in a very short time. This data set is an asset that needs to be regulated because of the strength of the massive pool of data accumulated.13

From this data set collected by the insurer, AI could be used to analyze the longevity of dental restorations (such as fillings or crowns) and identify patterns that may indicate high-quality work. For example, machine learning algorithms can be trained to recognize patterns and characteristics, such as smooth surfaces and precise fits, and to identify deviations from these standards.

However, it is essential to note that the quality of dental work is a complex and multifaceted issue. Other factors, such as the skills and experience of the dentist, the materials used, and the patient’s oral health history, play a role in determining the quality of dental work, as does the patient’s comfort, the durability of the treatment, and the overall aesthetic result. Therefore, while AI-powered radiography can assist in assessing the quality, it should be used in conjunction with other evaluation methods. Nevertheless, an unintended consequence of using AI-powered systems could be having every dental radiograph taken by a dentist go through a second-party evaluation process for some level of quality assessment.


An unintended consequence of AI in dental radiography is that it could disrupt the legacy volume-based model, where service is based on a fixed price, and there is greater payment when more services are performed. This commodity approach is commonly used today whether the patient is insured or not. Instead, imagine a value-based model predicated on patient outcomes and improving efficiency by creating therapeutic packages and a digital infrastructure that integrates the clinical process. In this value-based model, the revenue or reimbursement is tied to the ability of the clinician to achieve patient outcomes through the evaluation of clinical results. For example, taking a radiograph after treatment would allow some level of evaluation of the quality of dental work through an AI system. The insurer could evaluate the clinical result in combination with other measures (net promoter scales) already being used. This type of evaluation could identify potential fraud and dental work below the standard of care and build brand confidence in the insurer through the identification of a verified network of clinicians.

Together, the speed, ease and minimal cost of such an evaluation might compel insurers to insist clinicians send every pre and post-treatment radiograph for evaluation and design a new level of oversight. Moreover, the natural extension of such an arrangement creates yet another unintended consequence: When the insurer’s AI system concludes the patient’s treatment falls well below the standard of care or is potentially fraudulent, the concern could instantly be referred to dental regulators for an investigation. This could create a significant strain on the regulator’s resources.



The rapidly evolving nature of AI and emerging digital technologies will require adaptive action and ongoing monitoring from the Royal College of Dental Surgeons of Ontario (RCDSO) while also providing opportunities to improve patient safety further, redefine quality assurance and provide leadership through the development of dynamic guidelines and standards. Regulators will need to develop new risk assessment and management strategies which would be part of a continuous iterative process throughout the lifecycle of high-risk AI systems such as AI-powered dental radiography.

One key issue that must be addressed in regulating AI in dental radiography is the potential for AI to perpetuate or amplify existing biases and inequalities in healthcare. Therefore, regulatory bodies should also regulate the AI system to ensure that the system is transparent, accountable, and ethically responsible before deployment. This may include creating guidelines and standards for the development and use of AI such as the Beta principles for the ethical use of AI and data enhanced technologies in Ontario.14


If AI is used to automate specific tasks or decision-making processes, it may change the nature of dentists’ work and the skills and knowledge required to perform it. A potential approach to regulate this is through licensing and certification mechanisms. For example, the RCDSO could develop a certification program for dentists who wish to use AI in their practice, requiring them to meet specific standards and criteria to obtain and maintain their certification, as has been explored by the College of Physicians and Surgeons of Canada.15 This could ensure that only qualified and competent dentists use AI in their practice and provide a means of accountability for dentists who use AI responsibly and ethically.

There are also legal considerations that will need to be taken into account in the regulation of AI in dental radiography in Ontario. These include issues related to privacy, data protection, and liability. For example, regulatory bodies will need to consider the legal implications of the collection, use, and sharing of patient data by AI systems and the potential liability of dentists and other healthcare providers for errors or omissions made by AI.16


Liability for harm and accountability caused by AI-driven decisions is a significant issue raised with AI systems involved in clinical decision-making. Laws currently are unclear as to who will be held accountable for physical and emotional harm, privacy harm or discrimination harm caused by AI-driven decisions. As the people responsible for overseeing AI systems, it is obvious to hold dentists accountable in such situations. However, potential legal liability could not only be targeted toward dentists but also a larger group of stakeholders, including manufacturers, programmers and developers of the systems. Any AI-informed decision is a by-product and combination of the actions of all of the above groups; manufacturers and developers provide the initial algorithm design and training data. Dentists evaluate the AI-based results and finalize recommendations. If dentists were held solely responsible, they could refuse to use the technology. The professional liability program will need to guide dentists regarding their liability for using AI-powered radiography.


Another unintended consequence is that regulators could also audit dentists using AI-powered radiography to identify substandard dentistry. For example, evaluating a smooth restoration or endodontic obturation is possible today using existing AI radiography systems. However, as noted above, other factors will also play a role in determining the quality of dental work. Overall, while AI can provide some insights into the quality of dental work, a more holistic approach is needed to fully assess the quality of dental care, including the use of additional data sources and expert evaluation. Nevertheless, the ability to increase the breadth and speed of regulating clinicians is far from the intended purpose of AI dental radiography and could be a consequence too severe for dentists to use the technology.


New regulations will be required to ensure that AI-powered dental radiography is accurate, reliable, and fair while also protecting the privacy and security of patient data. It may also require the development of new laws or policies to address the unique challenges and risks posed by AI, such as tabled Bill C-27.17

Regulating AI technology is very different from regulating an intra-oral x-ray unit. If you buy an intraoral x-ray unit in Toronto and later relocate it to Ottawa, you know it will work the same in both places. Moreover, when another dentist uses the x-ray unit, it will operate like it always did. Algorithms, by contrast, change as human behaviour changes. As a result, algorithms resemble not the physical x-ray units regulated in the past, but something more like the bacteria in our intestines, living organisms that interact and evolve with us.18

To effectively regulate AI in dental radiography in Ontario, it will be necessary for regulatory bodies to adopt a proactive and forward-looking approach rather than a reactive and backward-looking one. This will require regulatory bodies to stay informed about the latest developments in AI and dental radiography. Therefore, an ongoing monitoring and development strategy to address the need for further recommendations in AI and emerging digital technologies is necessary and, fortunately, already in the works.19

To ensure that the needs and concerns of all stakeholders are taken into account, regulatory bodies should consult with a wide range of stakeholders, including healthcare providers, patients, and other interested parties, when developing regulatory frameworks for AI in radiography. This helps ensure that the technology is being used in a way that is responsive to the needs and values of society. Effective regulation of AI in radiography will require a balanced and nuanced approach. By following these principles, regulatory bodies can help ensure that AI is used ethically and responsibly in dentistry.



Many patients exist in a world of insufficient data, insufficient time, and insufficient context and are cared for by rushed or burnt-out healthcare providers. AI’s capabilities can return something invaluable to dental providers, which is the gift of time, resulting in a better patient experience. Moreover, AI dental radiography allows patients to benefit from more accurate diagnoses and develop more effective treatment plans. AI can assist dentists in communicating findings and helping patients understand their oral health condition. AI-based radiography can also streamline insurance claims processing. Overall, using AI in dentistry can improve the quality and efficiency of patient care and make it more accessible to a broader population.


The privacy and security of patient data must be protected in all stages of AI system design, starting from data collection and storage to the modelling and deployment of the system.20,21 At the data collection step, to ensure patient privacy and establish consent, clear information must be provided on how the data will be used.22 Although current privacy laws require consent for the collection, use and disclosure of personal health information, the consent is only intended for the specific group taking care of the patient’s health.16 With AI systems relying on patient data to “learn”, clarity is needed on whether consent for data sharing also applies to the AI system.23

A potential danger of using AI in radiography is the possibility of cyber-attacks or data breaches. AI systems rely on collecting, storing, and analyzing large amounts of sensitive patient data, which can be vulnerable to cyber attacks or data breaches.24,25 This could expose patients to identity theft or other forms of online fraud or compromise the privacy and security of their personal health information. To mitigate this risk, developers and healthcare providers must implement robust security measures to protect against cyber-attacks and data breaches and ensure that patient data is handled securely.26 Moreover, it needs to be clarified where the patient data is being analyzed (in Canada or beyond), complicating privacy concerns by moving patient data across borders.


AI-based radiography levels the playing field of the dentist–patient hierarchy and will fundamentally alter the relationship through the amplification and acceleration of knowledge transfer. Patients now have access to dental information that was historically only available to dentists. The dentist’s role could shift to curate evidence-based information in their new role as a guide or coach for the empowered patient, including the results of an AI radiography study. However, without clear regulations, we may only be steps away from an automated dental-x-ray kiosk at the shopping mall offering an alternative to a traditional dental appointment.


The AI system developers and manufacturers have disrupted dental radiography by amplifying and accelerating what already exists. They will derive the greatest immediate financial gain from the technology being used (Fig. 1) and, at present, leave the risk with the dentist. Insurers and patients will also significantly benefit without incurring additional responsibility. When technology moves faster than the law, the externalities or the harm it creates is not immediately obvious, and it needs to be determined on what basis the risks shift back to the creator of the technology.

Fig. 1

The potential relative shift of the benefits of AI-powered dental radiology among stakeholders.
The potential relative shift of the benefits of AI-powered dental radiology among stakeholders.

In the case of the developers and manufacturers, the risks need to be shifted back to them as they are in the best position to control the technology. Moreover, the transparency and explainability of the system should rest with them also, as they are custodians of the algorithms. With this, the cost of the potential harm should be absorbed by them to a greater extent. These potential harms include, among others, a loss of privacy, cyber security concerns, and tort and human rights issues. New standards and regulations need to address these potential harms.


With the significant potential of AI-powered dental radiography, marketed to assist dentists in providing more accurate procedures and diagnoses, dentistry will experience disruption and dislocation. This novel dental technology will allow earlier detection of dental disease and better clinical outcomes with unparalleled oversight. However, when you have a self-learning, self-actioning system that has unintended consequences, who takes responsibility? Early in a disruption, one can’t appreciate all the harms and shifting the risk is difficult. If the harm is viewed only from an economic focus, other externalities will be missed.

It stands to reason that the stakeholders developing, deploying or operating AI systems should be held accountable for their proper functioning. Machines will not replace dentists, but dentists using AI will soon replace those not using it, so it appears to be a shared responsibility requiring new regulatory frameworks. The choice is for regulators to take a leadership role in guiding that change and to plan for it or react to unforeseen and unintended explosive consequences.

Oral Health welcomes this original article.


  1. Siddique NH. Applied intelligence for industry 4.0. Boca Raton, FL: CRC Press,; 2023.
  2. Al-Sarem M, Al-Asali M, Alqutaibi AY, Saeed F. Enhanced Tooth Region Detection Using Pretrained Deep Learning Models. Int J Environ Res Public Health. 2022;19(22).
  3. Schwendicke F, Cejudo Grano de Oro J, Garcia Cantu A, Meyer-Lueckel H, Chaurasia A, Krois J. Artificial Intelligence for Caries Detection: Value of Data and Information. J Dent Res. 2022;101(11):1350-6.
  4. Hung KF, Yeung AWK, Bornstein MM, Schwendicke F. Personalized dental medicine, artificial intelligence, and their relevance for dentomaxillofacial imaging. Dentomaxillofac Radiol. 2023;52(1):20220335.
  5. Schwendicke F, Samek W, Krois J. Artificial Intelligence in Dentistry: Chances and Challenges. J Dent Res. 2020;99(7):769-74.
  6. Banting DW, Amaechi BT, Bader JD, Blanchard P, Gilbert GH, Gullion CM, et al. Examiner training and reliability in two randomized clinical trials of adult dental caries. J Public Health Dent. 2011;71(4):335-44.
  7. Valachovic RW, Douglass CW, Berkey CS, McNeil BJ, Chauncey HH. Examiner reliability in dental radiography. J Dent Res. 1986;65(3):432-6.
  8. Nguyen TT, Larrivee N, Lee A, Bilaniuk O, Durand R. Use of Artificial Intelligence in Dentistry: Current Clinical Trends and Research Advances. J Can Dent Assoc. 2021;87:l7.
  9. Bader JD, Shugars DA, Bonito AJ. Systematic reviews of selected dental caries diagnostic and management methods. J Dent Educ. 2001;65(10):960-8.
  10. Schwendicke F, Chaurasia A, Wiegand T, Uribe SE, Fontana M, Akota I, et al. Artificial intelligence for oral and dental healthcare: Core education curriculum. J Dent. 2023;128:104363.
  11. Norman CD, Skinner HA. eHEALS: The eHealth Literacy Scale. J Med Internet Res. 2006;8(4):e27.
  12. Mehta S, Grant K, Ackery A. Future of blockchain in healthcare: potential to improve the accessibility, security and interoperability of electronic health records. BMJ Health Care Inform. 2020;27(3).
  13. Doctorow C. Personal data is as hot as nuclear waste. The Guardian. 2008.
  14. Beta principles for the ethical use of AI and data enhanced technologies in Ontario. In: Ontario Go, editor.: King’s Printer for Ontario; 2023.
  15. Richard Reznick KH, Ranya Horsley, Mohsen Sheikh Hassani. Task Force Report on Artificial Intelligence and Emerging Digital Technologies. In: Canada RCoPaSo, editor. 2020.
  16. PIPEDA fair information principles. In: Canada OotPCo, editor. 2019.
  17. Canada Po. BILL C-27: An Act to enact the Consumer Privacy Protection Act, the Personal Information and Data Protection Tribunal Act and the Artificial Intelligence and Data Act and to make consequential and related amendments to other Acts: House of Commons; 2022 [Available from:
  18. Applebaum AP, P. How to put out democracy’s dumpster fire. The Atlantic. 2021 Apri 2021.
  19. Ontario RCoDSo. 432nd Meeting of Council. Toronto, Ontario; 2022 September 15, 2022.
  20. Ontario RCoDSo. Guidelines: Electronic Records Management. Toronto2012.
  21. PIPEDA Fair Information Principle 7- Safeguards. In: Canada OotPCo, editor. 2021.
  22. Guidance on inappropriate data practices: Interpretation and application of subsection 5(3). In: Canada OotPCo, editor. 2018.
  23. An Act to enact the Consumer Privacy Protection Act, the Personal Information and Data Protection Tribunal Act and the Artificial Intelligence and Data Act and to make consequential and related amendments to other Acts, 1st Sess. (2022).
  24. James Kaplan CT, Adam Tyra. Perspectives on transforming cybersecurity. McKinsey & Company; 2019.
  25. Morin P. Cybersecurity in Canada: Evolving threats and practical defences. Grant Thornton; 2021.
  26. Emily Mossburg JG, Hector Calzada. Beneath the surface of a cyberattack A deeper look at business impacts. Deloitte Touche Tohmatsu Limited; 2016.

About the Author

Dr. Peter Fritz

Dr. Fritz is a periodontist, scientist, mentor, and adjunct professor at three universities. He completed an MBA and then recently a law degree focusing on blockchain, cybersecurity and artificial intelligence. Peter lives by his academic mission: “Never stop learning because the world around you never stops teaching.” The author thanks Professor Dera J. Nevin at the University of Toronto Faculty of Law GPLLM program for inspiring this paper.