You must be signed in to read the rest of this article.
Registration on CDEWorld is free. Sign up today!
Forgot your password? Click Here!
While no technological advancement today is more certain to transform the future of dentistry than artificial intelligence (AI), probably none is less well understood by the people it is poised to benefit. AI is really not so mysterious. Unlike many dramatic portrayals of AI in the movies and media, it is not a machine that has been sent from the future to destroy people or refuse to "open the pod bay doors." Nor is it a computer whose intelligence is indistinguishable from that of a human being. That kind of AI, called "general AI," is still a fantasy. The AI that will transform dentistry is "narrow AI." It is sharply focused on a single task and extremely good at performing that task; so good, in fact, that it is able to operate at a level that matches expert human performance in quality and far surpasses it in speed. Moreover, it uncomplainingly accepts tasks that most people do not want to do, get tired of doing, or make mistakes doing.
Unlike the frightening machines of science-fiction fantasies, narrow AI is trustworthy and honest. This is why people are increasingly putting their faith in AI and handing over to it many tasks such as dictation, scheduling, sorting, labeling, route planning, and more, chores that traditionally have consumed a large percentage of one's workday. AI is now giving people back much of that time, allowing them to do more of the things that take real human intelligence and do those things better.
Deep Learning Process
Narrow AI systems develop their intelligence much the way biological brains do. A newborn baby's brain enters the world a blank slate, then gradually sorts sensory inputs into patterns that become a framework for understanding the world. Since there are all sorts of patterns in everything-some meaningful, some not-AI needs guidance about what kinds of patterns to look for, just as a baby does. For AI, the guidance comes in the form of data that has been "labeled" or "annotated" to draw attention to relevant patterns.1 Training an AI with data curated by human intelligence, a process called "deep learning," enables it to identify patterns with a human level of accuracy.
Humans interact with deep learning AI systems many times a day. The most familiar variety people encounter is computer vision: AI that can identify, track, and interpret visual imagery. If a computer vision system is fed enough images that are annotated with precise information defining what they depict, the system can learn to recognize even the most indistinct visual elements. Because these systems detect very slight variations between individual pixels, they can detect visual features that may escape the human eye, for example, the slightest sign of incipient caries in a dental x-ray.2 But what makes computer vision systems so valuable is their ability to easily handle huge volumes of visual data. These systems can deliver thousands of evaluations in the time it takes a human to deliver just one, and they do so while retaining perfect recall of every image they have ever seen.
Other types of AI systems understand and interpret human language. Almost every time someone runs a Google query, natural language processing is at work.3 This is also true when one uses autotranslation tools or asks a smart assistant to play their favorite song. Natural language processing and computer vision often run in tandem with data and predictive analytics systems, which permeate people's Facebook feeds, Netflix home screens, and Amazon recommendations. These systems use AI that is trained to find patterns or make predictions based on the trail of data people leave in their wake while searching for things, purchasing items, or seeking entertainment. Predictive analytics systems monitor personal spending habits to stop fraudulent charges. They track browsing behavior to drive digital advertising and guide commuters home on back roads to avoid heavy traffic.
Deep learning algorithms are ubiquitous, continuously learning from people, and delivering benefits large and small. Healthcare is no exception. As early as 2011, researchers from New York University Langone Health found that computer vision could find specific lung nodules up to 97% faster than a panel of radiologists.4 In hospitals over the past decade, computer vision has been deployed in a number of medical imaging tasks,5 including cancer detection and staging, lung segmentation,6 predicting therapeutic outcomes of depression,7 cardiovascular medicine,8 detection and classification of brain diseases,9 and many others. While the AI systems performing these functions today are often able to analyze and interpret medical imagery with accuracy superior to that of human clinicians, they are almost exclusively indicated for use as diagnostic aids to dental practitioners rather than as stand-alone diagnostic devices. Importantly, when in that assistive capacity, AI significantly elevates a clinical practitioner's diagnostic performance. More accurate diagnosis reduces costly treatment redirects and eliminates repetitive routine tasks, including third- and fourth-opinion consults and associated redundant imaging.
Bringing Consistency to Dentistry
In dentistry, it is anticipated that AI will bring similar efficiencies as those seen in other areas of healthcare, and is likely to bring them speedily, for several reasons. Firstly, the entrepreneurial and consumer-oriented nature of the dental profession can help facilitate more agile and frictionless AI adoption across private dental practices, potentially fueled by patient demand. Secondly, the narrow focus of care upon a single part of the body-namely, the oral cavity and/or associated areas-simplifies the technical challenge of developing AI capabilities that are broadly applicable across dental medicine. Thirdly, compared to many medical fields, the relatively lower risk of harm resulting from dental treatment reduces the risk associated with the use of AI systems in dental care, lowering the stakes of dental AI adoption for patients, doctors, and medical regulators alike. Finally, the rising rate of practice consolidation in dentistry is increasing the need, as well as demand, for new tools and methods for ensuring clinical consistency, operational efficiency, and liability reduction across centrally managed but operationally decentralized clinical facilities. Combined, these key factors make dentistry a fertile ground for AI to plant its roots and quickly deliver fruit. Figure 1 through Figure 5 depict various ways in which AI can be utilized in dentistry.
The most significant effect of AI on patient care may indeed be consistency. Patients are sensitive to inconsistency in medical diagnosis and treatment and develop lasting aversion to it when they notice it. This presents a challenge for dentistry, which has a well-documented inconsistency problem.
Many dental providers may recall a 1997 Reader's Digest report by an investigative journalist who visited 50 different dentists across the United States, showed each the same x-rays of his teeth, and received 50 different diagnoses and treatment plans ranging in price from under $500 to more than $29,000.10 One could hardly have found a more striking confirmation of the often capricious nature of dental diagnosis.
More formal studies have confirmed this impression. A study recently conducted by the Dental AI Council asked 136 licensed dentists to provide diagnosis and treatment plans based on review of a full-mouth x-ray image set.11 The respondents never delivered better than 50% diagnostic agreement on any given tooth and proposed full-mouth treatment plans with costs ranging from $300 to $36,000. Even dropping the outliers, the range of responses and the lack of tight clustering around the mean was startling. The differences could be attributed to any number of factors, whether training, experience, skill, state of mind, business motives, or many others. But whatever the cause, it is clear that inconsistency is endemic in dental diagnosis.
Computer vision systems, on the other hand, are consistent. They will perform the same job the same way every time, because consistency is intrinsic to the deterministic nature of their intelligence. Their diagnostic performance is free of the preconceptions, biases, distractions, and fatigue that affect even the most expert human radiologist. They equal the performance of expert readers in head-to-head diagnostic challenges and consistently elevate diagnostic accuracy when used as a detection aid.12
Speed, Intelligence, and Efficiency
AI is very fast. A computer vision system can analyze thousands of x-rays in minutes, allowing a dental office to rapidly evaluate every historical x-ray in its practice management software (PMS) in order to identify missed diagnoses and gain insight into its diagnostic strengths and weaknesses. Instantaneous processing of a single file, on the other hand, provides practitioners with chairside second opinions with authoritative quality that delivers cascading benefits to treatment planning, patient trust, insurance compliance, and medical liability. As patient trust increases, it is logical to think that so, too, will patient willingness to seek and accept dental care. Better population-wide oral health, in turn, will likely follow, as will a stronger, more trustworthy dental industry.
The intelligence that AI brings to practice management can also be applied to human resources, procurement, and administration. These efficiencies will be particularly significant for larger groups, where the benefits of patient need-driven staffing, marketing, and inventory management apply at enterprise scale.
Seeds Still Being Sown
Dentistry is only at the beginning of its AI revolution. The seeds of AI's biggest benefits to dentistry are still being sown. These seeds are fertilized by data-data that is multiplied across tens of thousands of individual practices, dental schools, laboratories, insurance carriers, PMSs, original equipment manufacturers, materials suppliers, distributors, and so on.
With these massive population-level data sets feeding machine learning systems, the benefits to public health are likely to expand in ways that cannot yet even be clearly foreseen. AI's ability to rapidly sift through data for nuggets of insight will allow it to find correlations between symptoms and trends within subgroups that will indicate the need for further study. Insights into lifestyle, genetic, and systemic aspects of oral health may add a holistic dimension to oral health that has been invisible to practitioners until now. The dentist, then, may become something more than a dentist.
Recognizing the power of AI is a good starting point for dentists. Beyond that, understanding its strengths and weaknesses will then allow dental providers to approach the technology responsibly. But more importantly, this understanding will show practitioners how to nurture AI's strengths in ways that amplify its future utility.
The Importance of Recording Data
Because AI's strength grows from data, dentists should make every effort to record data on everything they can. They should institute processes for tracking more of what they are doing in the name of dentistry-even if it is inconvenient at first. Narrow AI systems can make use of data in ways that humans' general intelligence brains simply cannot. Thus, although dentists may not yet fully understand the value of the data they create, they can let themselves be guided by their understanding of the systems that can understand it. By striving to save practice data, and also seek ways to make the data public while protecting patient privacy, dental providers will allow the future of the profession to benefit from the actions of today.
Acknowledgment
The images presented in Figure 1 through Figure 5 are courtesy of Pearl Inc. and used with permission.
About the Author
Kyle Stanley, DDS
Private Practice, Beverly Hills, California
Queries to the author regarding this course may be submitted to authorqueries@broadcastmed.com.
References
1. Thrall JH, Li X, Li Q, et al. Artificial intelligence and machine learning in radiology: opportunities, challenges, pitfalls, and criteria for success. J Am Coll Radiol. 2018;15(3 Pt B):504-508.
2. Lee JH, Kim DH, Jeong SN, Choi SH. Detection and diagnosis of dental caries using a deep learning-based convolutional neural network algorithm. J Dent. 2018;77:106-111.
3. Karakulah G, Dicle O, Kosaner O, et al. Computer based extraction of phenotypic features of human congenital anomalies from the digital literature with natural language processing techniques. Stud Health Technol Inform. 2014;205:570-574.
4. Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning. Nat Med. 2018;24(10):1559-1567.
5. Savadjiev P, Chong J, Dohan A, et al. Demystification of AI-driven medical image interpretation: past, present and future. Eur Radiol. 2019;29(3):1616-1624.
6. Hu Q, de F Souza LF, Holanda GB, et al. An effective approach for CT lung segmentation using mask region-based convolutional neural networks. Artif Intell Med. 2020;103:101792.
7. Lee Y, Ragguett RM, Mansur RB, et al. Applications of machine learning algorithms to predict therapeutic outcomes in depression: a meta-analysis and systematic review. J Affect Disord. 2018;241:519-532.
8. Krittanawong C, Zhang H, Wang Z, et al. Artificial intelligence in precision cardiovascular medicine. J Am Coll Cardiol. 2017;69(21):2657-2664.
9. Khan P, Kader MF, Riazul Islam SM, et al. Machine learning and deep learning approaches for brain disease diagnosis: principles and recent advances. IEEE Access. 2021;9:37622-37655. doi: 10.1109/ACCESS.2021.3062484.
10. Ecenbarger W. I went to 50 different dentists and almost all of them gave me a different diagnosis. Reader's Digest. February 1997. Updated November 22, 2022. https://www.rd.com/article/how-honest-are-dentists/. Accessed March 7, 2023.
11. Dental AI Council (DAIC). Inconsistency in Radiographic Dental Diagnostics & Treatment Planning. December 2020. https://static1.squarespace.com/static/5f23743e90bfa971db0fe5b1/t/5fe0ccb786144b2be1731386/1608568403250/DAIC+-+TreatmentStudy+-+12.16.2020.pdf. Accessed March 7, 2023.
12. Devito KL, de Souza Barbosa F, Felippe Filho WN. An artificial multilayer perceptron neural network for diagnosis of proximal dental caries. Oral Surg Oral Med Oral Pathol Oral Radiol Endod. 2008;106(6):879-884.
