Artificial intelligence in dentistry – where are we?

Artificial intelligence (AI) and machine learning (ML) are already part of our everyday lives. Most of us are familiar and comfortable with using smart phones and with applications such as Google Maps, facial recognition, predictive text and digital voice assistants (e.g. Siri, Alexa and Cortana). We are also likely to appreciate advances in medical technology, especially when those improvements lead to early diagnoses and interventive treatment for our families and ourselves. But what about when it comes to dentistry? Did you know that in July 2024, an AI-guided robot performed the first crown preparation on a live human subject?

Does this mean that dentists will, in future, become redundant? Or, is the field of robotics and AI providing opportunities for practitioners to offer enhanced precision to ultimately raise the overall standard of care, and perhaps also decrease costs?

What is AI? The term “artificial intelligence” (AI) refers to the idea of machines with advanced computer programming (algorithms) being capable of performing tasks which would normally require human intelligence. If this computer intelligence is learned by the machine from datathen this is termed machine learning (ML). Computers may utilise large quantities of “input” data to then “learn” to recognise patterns and so to formulate predictions. A common type of machine learning software is called a neural network, which has layers of ‘neurones’. Complex input data such as from images is fed into the neural network which then learns to recognise features such as decay in a dental radiograph.

Historic applications: Machine learning is not just restricted to image recognition. For many years now, practitioners have been utilising AI for dictation – such as when writing reports and clinical notes – where the program precisely converts voice to text. More recently, generative AI programs based on a newer type of neural network (called a transformer) have developed more advanced functionality such as predictive and summarising capabilities. AI scribes can even capture speech that is taking place during a clinical encounter, then convert the audio data into text. It is worthwhile noting that all existing regulations (such as the Australian Privacy Principles (APP’s)) and data protection principles must be adhered to in the collection, use, disclosure and storage of information collected this way. In SA, the Surveillance Devices Act2016 | South Australian Legislation governs the installation, use and maintenance of listening and surveillance devices. The practical effect of the legislation centres on the rights of individuals to privacy and requires the patient to provide specific consent whenever they are being recorded. The patient must understand what is being recorded, how/when the information will be used/retained/shared and whether the resulting recording will be stored securely as part of their clinical record or destroyed. It is incumbent upon the practitioner to gain valid consent before any recording takes place. At the conclusion of the appointment, it is also important for the practitioner to also take the time to customise/correct the output data (clinical record) – or risk the record being inaccurate or invalid. In this regard, the treating practitioner is liable for any errors within the patient record – even where those errors are generated by an AI scribe.

Current applications of AI in dentistry also include in diagnosis, treatment planning and monitoring of patient progress. The uptake of diagnostic applications are prevalent in the field of orthodontics where AI can assist with cephalometric analyses and reliably forecast the need for tooth extraction and/or orthognathic surgery. CAD/CAM design and manufacturing of orthodontic aligners is now common as is remote patient monitoring. In periodontology, AI algorithms have been used in patient assessment to identify changes in soft tissues, bone density, alveolar bone loss, and to classify the stages of periodontitis on radiographs.Similarly, in restorative dentistry, AI analysis of intra-oral radiology can assist in the early detection, diagnosis and management of dental caries and CAD/CAM is used in construction of indirect restorations. In endodontology, AI contributes to the identification of periapical lesions, root and canal morphology, estimation of working length, root fractures, and treatment prognosis. In oral and maxillofacial surgery, AI algorithms are employed in diagnosis, preoperative surgery planning – particularly in the assessing the complexity and predicted outcome of impacted third molar surgery. In implantology, AI systems are being employed in analysis of CBCT data, virtual treatment planning, visualization of treatment outcomes and guided placement of instruments.

What’s next? Whilst AI is already being employed in multiple areas of dental clinical practice, it is worthwhile remembering that the applications are “tools” and that whilst their capability and accuracy is impressive, their field of operational capacity is necessarily narrow. For example, one form of AI can be employed to analyse dental radiographs (BWS, IOPAs and OPGs) to detect caries, calculus, periodontal disease, periapical radiolucencies, cysts and bone lesions with a high degree of accuracy. But the program’s ability stops there. The algorithm cannot yet conduct a visual inspection of the dentition or utilise an explorer or periodontal probe. It cannot do tooth percussion or palpation tests. It does not incorporate additional factors into diagnoses such as whether the patient is at high risk of disease progression due to dietary or lifestyle factors. The program simply alerts the clinician to the likelihood of a particular diagnosis for a particular area/tooth. Clinical oversight and input are still required to consider the implications of the x-ray findings and to advise patients on risks and appropriate treatment options.

It is clear that AI technology is here to stay, and in future, it is likely that more and more sophisticated tools will be developed. AI can already learn from multiple information sources (multi-modal data) and is capable of diagnoses beyond human capacity. For example, optometrists can take photographs of the retina using specialised equipment (a fundus camera). The photographic image can be analysed by an AI algorithm incorporating other medical data to reliably predict if the patient is at risk of heart disease.

Regulation and ethical considerations: Dental AI teaching hubs concentrating on delivery of a digital workflow in dentistry are already available in New Zealand. Similar centres for CPD learning are being developed in Australia too. Whilst the capability of the new technology is enticing, patient safety and the delivery of good clinical outcomes are paramount in dental clinical practice. The ADA has prepared a policy on AI which cautions against over-reliance or indiscriminate use of technology and recommends use should be restricted to trained clinicians (ada_policy-6-34_artificial-intelligence-in-dentistry.pdf):

Generative AI systems such as ChatGPT can authoritatively and convincingly produce false, inappropriate, and dangerous content, often without the ability for verification or validation. If Generative AI systems are used in clinical decision-making by laypersons, or for clinical decision-making at scale, action taken based on unverifiable responses has the potential to cause patients’ harm.

AHPRA, too, has issued guidance including describing the overarching principles to be applied when employing AI tools in a healthcare setting: Australian Health Practitioner Regulation Agency - Meeting your professional obligations when using Artificial Intelligence in healthcare