Harnessing the Power of Large Language Models in Orthodontic Practice

Advancements, Challenges, and Applications

I. Introduction

In recent years, the rapid advancements in artificial intelligence (AI) have sparked considerable interest and debate within the healthcare community. Generative Pre-trained Transformer Networks (GPTN) and Large Language Modeling (LLM) are among the most prominent AI technologies that have shown promise in revolutionizing various aspects of healthcare, including diagnostics, treatment planning, and patient education. While these AI applications have demonstrated considerable potential, it is essential to approach their implementation in healthcare settings with caution, taking into account the ethical, technical, and practical implications of their use.

One of the central concerns surrounding the integration of GPTNs and LLMs in healthcare is the fear that AI will replace human professionals, a notion that has gained traction in the popular imagination. However, it is vital to challenge this perception and recognize that AI technologies, when appropriately implemented, can act as powerful complementary tools for orthodontic practitioners and other healthcare professionals, enhancing their capabilities and expertise rather than displacing them. The integration of AI into orthodontic practice has the potential to create an improved standard of care by promoting greater practice efficiencies, streamlining diagnostic and treatment processes, and allowing practitioners to dedicate more time to patient care. This, in turn, can contribute to higher quality patient outcomes and increased satisfaction with orthodontic treatment, such as reduced treatment time and improved accuracy of diagnosis.

On the other hand, there are potential pitfalls of incorporating AI technologies into orthodontic practice, such as the risk of overreliance on AI systems, the potential for bias in data-driven decision-making, and the need for robust data security protocols to protect patient privacy. By taking these potential risks into account, orthodontic practitioners can ensure the responsible and ethical use of AI to ensure optimal patient care and the continued growth and development of the orthodontic profession.

In this paper, we will provide a thorough and cautious exploration of the key areas of LLM integration in orthodontics, including diagnosis and treatment planning, individualized patient education, and customized orthodontic appliances. We will discuss the current state of LLM applications in healthcare, the challenges and limitations associated with their implementation, and the potential benefits and pitfalls of incorporating AI technologies into orthodontic practice. By doing so, we aim to offer a balanced and informed perspective on the transformative potential of GPTNs and LLMs in the field of orthodontics, while emphasizing the importance of responsible and ethical use of AI to ensure optimal patient care and the continued growth and development of the orthodontic profession.

II. Overview of Literature on LLMs and GPT in Medicine and Dentistry

Natural Language Processing (NLP) models are a type of artificial intelligence, specifically advanced machine learning models, designed to understand and generate human-like text. These models are developed by training them on vast amounts of textual data from diverse sources, enabling them to capture intricate patterns and nuances in language. As a result, NLP models can perform various language-based tasks, such as answering questions, summarizing text, and generating sentences in a coherent and contextually appropriate manner. Generative Pre-trained Transformer (GPT) algorithms are a class of Natural Language Processing (NLP) models that excel in natural language understanding and generation. They employ a unique architecture called the Transformer, which facilitates parallel processing of textual data and enhances the learning of long-range dependencies within text. The pre-training phase is followed by fine-tuning on specific tasks, enabling GPT-4 algorithms to perform various language-based functions such as translation, summarization, and question-answering with high accuracy. The power of NLP models lies in their ability to process and analyze complex information efficiently, which makes them well-suited for applications in medicine, dentistry, and other fields where expert knowledge is required. For instance, NLP models have been used to identify and diagnose rare diseases, detect medical conditions from patient images, and predict the effectiveness of treatments. In addition, NLP models have been used to parse electronic health records (EHRs), assist with patient triage, and enhance patient education. In dentistry, NLP models have been used to detect caries and periodontal diseases from dental images.

Studies validating NLP models against the current standard of medical decision making (USMLE) demonstrate their capacity to assist healthcare professionals. Strong conclusions cannot be drawn from hypothetical testing, but it is important to see whether these results translate into clinical practice. In addition to achieving top-tier results on the USMLE, GPT-4 has been used in former real medical scenarios to diagnose rare conditions. Dr. Isaac Kohane, Chair of the Department of Biomedical Informatics at Harvard Medical School, has been using GPT-4 for several months and testing its abilities against old cases in his soon-to-be published book, The AI Revolution in Medicine: GPT-4 and Beyond [1]. Other practical applications have been proposed, such as building causal graphs, parsing electronic health records (EHR), assisting with patient triage, and enhancing patient education.

In the realm of causal graph construction, research suggests that while natural language processing models (NLP) have potential for constructing directed acyclic graphs (DAGs), their accuracy is contingent on the language used to describe the relationships [2,3]. In the fields of medicine and dentistry, DAGs play a crucial role in understanding and modeling complex causal relationships between various factors, such as genetic, environmental, and behavioral variables. They are especially valuable for elucidating causal pathways, disentangling confounding factors, and identifying potential targets for interventions. By visualizing and analyzing the relationships among variables using DAGs, researchers and practitioners can gain essential insights that help guide the development of effective diagnostic tools, therapeutic strategies, and preventive measures to improve patient care and health outcomes. This suggests the expert, in this case an orthodontist, still has a major role in the success of NLP implementation in orthodontic clinics. An NLP begins to be as effective as the training data it is fine-tuned on.

Moreover, NLP can parse electronic health records (EHR) information efficiently, as shown by Yang et al., enabling more accurate patient information extraction, cohort identification, and pharmacovigilance [4]. Although orthodontics is not a heavily pharmacologically based profession, extraction of patient information from written, clinical, and radiographic information is vital for accurate diagnosis. Sezgin et al. demonstrate that NLP can also streamline patient triage through chatbot integration, reducing overhead at clinics and enhancing the quality of care [5]. An orthodontic clinic can run more efficiently if the clinical team is given information about an orthodontic complication such as a broken bracket, therefore NLP provide an opportunity to prepare the clinical team quickly in real-time through a chatbot. Additionally, NLP can improve patient education by providing interactive computer-aided diagnosis, as illustrated by Wang et al., enabling patients to better understand their symptoms, diagnoses, and treatment options [6].

In dentistry, NLP such as GPT-3 offer multiple benefits, as documented by the 2023 article "ChatGPT for Shaping the Future of Dentistry" [7]. These include efficient text mining from unstructured data, streamlined documentation tasks, data-fused diagnosis, and personalized treatment plans. Eggmann et al. further highlights the potential of NLP in dental medicine for clinical decision support, text summarization, efficient writing, automated appointment scheduling, and multilingual communication [8]. NLP can also be used to create virtual assistants that can provide personalized advice and recommendations to patients, as well as to automate administrative tasks such as billing.

However, integrating AI into healthcare is not without challenges. Sezgin et al. and Eggmann et al. emphasize concerns related to processing needs, data confidentiality, cybersecurity, model biases, evaluation metrics, and establishing acceptable usage boundaries in scientific writing [5,9]. Additionally, ensuring the accuracy and security of health-related information provided by NLP is vital to protect against misinformation and maintain patient data confidentiality.

In conclusion, NLP have the potential to significantly impact orthodontics and healthcare at large. While their applications are promising, it is crucial to address the associated challenges and limitations to ensure the safe, ethical, and effective integration of AI into these fields.

III. LLM-assisted diagnosis and treatment planning

The potential of artificial intelligence (AI) to improve orthodontic diagnosis and treatment planning is increasingly evident. With the integration of large language models (LLMs) such as GPT into imaging software and electronic health records, orthodontists can access an advanced, comprehensive array of tools to aid in the decision-making process. AI and machine learning algorithms have been integrated into imaging and treatment planning software, most famously in CBCT analysis and clear aligner software. Although AI is related to LLMs, AI is not the focus of this study and the reader should instead read the following cited articles as primers on artificial intelligence and machine learning in orthodontics [10,11,12].

One significant benefit of LLM-integrated diagnosis and treatment is the ability to generate precise treatment plans. LLMs and GPTs can organize diagnostic information via specific commands and instructions, provided that orthodontists offer clear, explicit instructions based on clinical examination, radiographic imaging, and patient history. This approach ensures that AI-generated insights remain closely aligned with the orthodontist's expertise and clinical judgement. LLMs can output language that is clear, concise, and with multiple styles that may be more applicable to the orthodontist or the patient. Since LLMs are sensitive to inputs, it is imperative that the orthodontist is well versed in orthodontic theory and practice when fine-tuning a customized LLM.

Another advantage of incorporating AI into orthodontic practice is the reduction in treatment planning time. LLMs and GPTs can be customized for each orthodontist according to their supplies, philosophy, and preferences, thereby streamlining the process of filtering data from various sources to synthesize treatment plans. Orthodontists are not comfortable with all possible tools available but are constrained by the availability of certain appliances and comfort with certain procedures and techniques. Therefore, LLMs can be sensitive to these preferences and create treatment plans that are individually customized for each orthodontist’s abilities. Consequently, this accelerates the planning process, improving clinical efficiency and patient satisfaction.

AI-assisted diagnosis and treatment planning can also contribute to better patient outcomes. By leveraging AI-generated insights, orthodontists can develop more personalized treatment plans, enhancing the precision and predictability of treatment results. LLMs are based on training data, therefore if previous treatment information is used as the basis, an orthodontist is better able to understand his previous results and capabilities and fine-tune his treatment mechanics and intervals. This ultimately leads to more successful treatment outcomes and higher levels of patient satisfaction. Additionally, the orthodontist can explore new and creative treatment mechanics by parsing through LLM generated treatment combinations. An orthodontist is limited by his time and working memory, but an LLM can have an almost infinite amount of scientific literature and case history to be trained on. Therefore, the combinations created by a GPT and LLM are exponentially higher. Yet, it is still up to the orthodontist to decide whether the proposed ideas are viable or not.

One practical application of AI-generated insights in orthodontics could involve a web-based interface or application, which orthodontists can use to access AI-generated insights from previous patient treatments that resemble the current patient's case. This would facilitate case review, particularly in large practices with access to extensive patient data. Moreover, LLMs and GPTs could generate creative treatment mechanics and ideas by synthesizing previously completed treatment mechanics, offering orthodontists novel solutions they may not have considered otherwise. However, the ultimate responsibility still lies with the orthodontist to determine the feasibility of these proposed treatment mechanics.

In summary, AI-assisted diagnosis and treatment planning, utilizing LLMs such as GPT, offer numerous benefits for orthodontic practice, including precise treatment plans, reduced planning time, and better patient outcomes. By developing web-based interfaces or applications, orthodontists can leverage AI-generated insights to improve their clinical decision-making and devise innovative treatment mechanics, ultimately enhancing patient care and the overall orthodontic experience.

IV. Customized patient education

The implementation of large language models (LLMs) and GPT in orthodontics has the potential to significantly improve customized patient education. The development of chatbots for patient education systems, has shown to be effective in improving hygiene compliance among adolescent orthodontic patients [13]. Yet, there are no studies which have integrated LLMs for delivery of customized orthodontic instruction and education. As clear aligners become more popular and the preferred treatment choice of the patient, compliance becomes a more important variable for the orthodontist [14,15]. The needs for a compliance-based education and reminders individualized for patients may become a valuable tool for the orthodontist. More traditional orthodontic auxiliaries such as headgear, elastics, and palatal expanders requiring screw-turning also require compliance from the patient [16,17,18]. Throughout orthodontic treatment, it becomes increasingly important to ensure patients understand and adhere to their prescribed treatment regimens.

Customized patient education can be tailored to each patient's needs, such as the frequency of alerts, language preferences, and the specific compliance-based appliance for each patient. This personalized approach is more effective than providing a generic sheet of paper with instructions, or even an app that delivers instructions on a pre-determined timetable. Other advantages of customized patient education include tailored educational content delivered through a patient portal, email, or mobile app, which allows for more targeted and engaging communication with patients. Ideally, patients will be able to interact with a chatbot to receive only pertinent information related to their treatment. Additionally, the patient would be able to add their own instructions to the chatbot for reminders based on their own preferences. This way, both the orthodontist and the patient have influence on education and reminders, instead of one-sided control. Furthermore, the customization of patient education enhances patient satisfaction, as patients feel better informed and more involved in their treatment process. Overall, the use of LLMs and GPT for customized patient education in orthodontics has the potential to greatly improve patient understanding, compliance, and satisfaction, ultimately contributing to more successful treatment outcomes.

V. LLM Integration into CAD/CAM-designed orthodontic appliances

The evolution of 3D printing and computer-aided design and computer-aided manufacturing (CAD/CAM) technologies in orthodontics has significantly impacted the field, providing numerous advantages to practitioners and patients alike [19]. Orthodontists have been employing software such as Meshmixer, Deltaface, 3Shape Appliance Designer, and OnyxCeph for over a decade, since at least 2010, to create highly individualized appliances for their patients [20]. This technology is far from novel and has been extensively incorporated into orthodontic practice, offering an approach to treatment planning and execution that is efficient and viable [21]. The use of CAD/CAM and 3D-printed orthodontic appliances offers a multitude of benefits, including the ability to customize appliances to meet the unique needs of each patient, improved treatment efficiency, and enhanced patient comfort [22,23].

The integration of large language models (LLMs) into these advanced technologies promises to further streamline and optimize the fabrication process for customized orthodontic appliances. LLM-integrated, AI-driven digital models enable orthodontists to save precious time by dictating instructions for appliances, rather than laboriously designing them from scratch. This innovative approach allows practitioners to focus on refining and perfecting the final design, ensuring the highest level of patient care. Spline, a web-based, 3D design tool, exemplifies the potential of LLM integration, as it incorporates LLMs into its AI, permitting designers to use written prompts to generate, edit, and brainstorm 3D designs (24,25). Other CAD/CAM technologies can adopt a similar strategy, integrating LLMs and GPT via text-to-design modules for orthodontic appliances. By embracing the capabilities of LLMs in CAD/CAM-designed orthodontic appliances, the orthodontic community can continue to advance and deliver exceptional patient outcomes, building upon the already established foundation of CAD/CAM technologies in the field.

VI. Conclusion

In conclusion, the integration of GPT and large language models (LLMs) into orthodontic practice holds tremendous potential for enhancing the field in several key areas. This paper has explored the application of GPT and LLMs in medicine and dentistry through a thorough analysis, highlighting the promising developments and possibilities these technologies offer. We have examined the benefits of LLM-integrated diagnosis and treatment planning, which include more precise treatment plans and predictions, as well as the ability to customize patient education. This customization can help to improve patient understanding of treatment plans, potentially increase compliance, and result in higher patient satisfaction.

Furthermore, the integration of LLMs into CAD/CAM-designed orthodontic appliances is poised to enhance the way orthodontists create and adapt appliances for individual patients, ultimately leading to greater efficiency and enhanced patient comfort. The integration of orthodontic practice with cutting-edge technologies, such as GPT and LLMs, can assist in the progression of the field, providing practitioners with innovative tools that save time, reduce costs, and ultimately lead to improved patient experiences and outcomes. For example, LLMs can be used to generate personalized treatment plans that are tailored to each patient's individual needs, as well as to create patient-specific educational materials that can help to improve their understanding of the treatment process. As research continues to unveil the full potential of GPT and LLMs in orthodontics, it is crucial for the orthodontic community to embrace these cutting-edge technologies and integrate them into daily practice to maintain the highest standard of care and deliver the best possible results for patients.

References

1.     Lee, Peter, Carey Goldenberg, and Isaac Kohane. The AI Revolution in Medicine: GPT4 and Beyond. 1st ed., Pearson, 13 May 2023.

2.     Long, Stephanie et al. “Can large language models build causal graphs?” ArXiv abs/2303.05279 (2023): n. pag.

3.     Arsenyan, Vahan and Davit Shahnazaryan. “Large Language Models for Biomedical Causal Graph Construction.” ArXiv abs/2301.12473 (2023): n. pag.

4.     Yang, X., Chen, A., PourNejatian, N. et al. A large language model for electronic health records. npj Digit. Med. 5, 194 (2022). https://doi.org/10.1038/s41746-022-00742-2

5.     Sezgin E, Sirrianni J, Linwood SL. Operationalizing and Implementing Pretrained, Large Artificial Intelligence Linguistic Models in the US Health Care System: Outlook of Generative Pretrained Transformer 3 (GPT-3) as a Service Model. JMIR Med Inform. 2022 Feb 10;10(2):e32875. doi: 10.2196/32875. PMID: 35142635; PMCID: PMC8874824.

6.     Wang, Sheng et al. “ChatCAD: Interactive Computer-Aided Diagnosis on Medical Image using Large Language Models.” ArXiv abs/2302.07257 (2023): n. pag.

7.     Huang, Hanyao et al. “ChatGPT for Shaping the Future of Dentistry: The Potential of Multi-Modal Large Language Model.” (2023).

8.     Eggmann F, Weiger R, Zitzmann NU, Blatz MB. Implications of large language models such as ChatGPT for dental medicine. J Esthet Restor Dent. 2023 Apr 5. doi: 10.1111/jerd.13046. Epub ahead of print. PMID: 37017291.

9.     Eggmann, Florin & Blatz, Markus. (2023). ChatGPT: Chances and Challenges for Dentistry. The Compendium of continuing education in dentistry. 44. 220-224.

10. Monill-González, A, Rovira-Calatayud, L, d’Oliveira, NG, Ustrell-Torrent, JM. Artificial intelligence in orthodontics: Where are we now? A scoping review. Orthod Craniofac Res. 2021; 24 (Suppl. 2): 6– 15. https://doi.org/10.1111/ocr.12517

11. Retrouvey JM. The role of AI and machine learning in contemporary orthodontics. APOS Trends Orthod 2021;11(1):74-80.

12. Hussain, M. A., et al. "Artificial Intelligence in Orthodontics: A Review." International Journal of Health Sciences, no. II, 17 May. 2022, pp. 9378-9383, doi:10.53730/ijhs.v6nS2.7445.

13. Zotti, Francesca et al. “Usefulness of an app in improving oral hygiene compliance in adolescent orthodontic patients.” The Angle orthodontist 86 1 (2016): 101-7 .

14. Orhan, C., Tuncay., S., Jay, Bowman., Jonathan, L., Nicozisis., Brian, D., Amy. "Effectiveness of a compliance indicator for clear aligners.." Journal of clinical orthodontics, 43 (2009).:263-.

15. Timm LH, Farrag G, Baxmann M, Schwendicke F. Factors Influencing Patient Compliance during Clear Aligner Therapy: A Retrospective Cohort Study. J Clin Med. 2021 Jul 14;10(14):3103. doi: 10.3390/jcm10143103. PMID: 34300269; PMCID: PMC8303492.

16. Al-Moghrabi, Dalya et al. “Compliance with removable orthodontic appliances and adjuncts: A systematic review and meta‐analysis.” American Journal of Orthodontics and Dentofacial Orthopedics 152 (2017): 17–32.

17. Egolf, R J et al. “Factors associated with orthodontic patient compliance with intraoral elastic and headgear wear.” American journal of orthodontics and dentofacial orthopedics : official publication of the American Association of Orthodontists, its constituent societies, and the American Board of Orthodontics 97 4 (1990): 336-48.

18. Veeroo, Helen & Cunningham, Susan & Newton, Jonathon & Travess, Helen. (2014). Motivation and compliance with intraoral elastics. American Journal of Orthodontics and Dentofacial Orthopedics. 146. 33–39. 10.1016/j.ajodo.2014.03.021.

19. Simon Graf, Nour E. Tarraf, Neal D. Kravitz, “Three-dimensional metal printed orthodontic laboratory appliances”, Seminars in Orthodontics, Volume 27, Issue 3, 2021, Pages 189-193, ISSN 1073-8746, https://doi.org/10.1053/j.sodo.2021.09.005

20. Wilmes B, Drescher D. Application and effectiveness of the Beneslider: a device to move molars distally. World J Orthod. 2010 Winter;11(4):331-40. PMID: 21490998.

21. Graf S, Vasudavan S, Wilmes B. CAD-CAM design and 3-dimensional printing of mini-implant retained orthodontic appliances. Am J Orthod Dentofacial Orthop. 2018 Dec;154(6):877-882. doi: 10.1016/j.ajodo.2018.07.013. PMID: 30477785.

22. Tsolakis IA, Gizani S, Tsolakis AI, Panayi N. Three-Dimensional-Printed Customized Orthodontic and Pedodontic Appliances: A Critical Review of a New Era for Treatment. Children (Basel). 2022 Jul 23;9(8):1107. doi: 10.3390/children9081107. PMID: 35892610; PMCID: PMC9332207.

23. Wilmes, Prof. Dr. Benedict and Prof. Dr. Dieter Drescher. “CAD-CAM workflows for palatal TAD anchored appliances.” Seminars in Orthodontics (2023): n. pag.

24. Spline AI - 3D Design Faster With AI. spline.design/ai.

25. Spline - Design Tool for 3D Web Browser Experiences. spline.design/#features.