Abstract

The promise of artificial intelligence (AI) and deep learning in medicine and oral-maxillofacial surgery should not be underestimated. As modern medical imaging transformed how clinicians visualize anatomy and pathology, the clinical application of AI in diagnostic and prognostic predictability may revolutionize standard of care.1Kelly C.J. Karthikesalingam A. Suleyman M. et al.Key challenges for delivering clinical impact with artificial intelligence.BMC Med. 2019; 17: 195Crossref PubMed Scopus (460) Google Scholar Today, AI has been used in the detection of cancerous lesions of the tongue, in the computation of facial attractiveness for cosmetic surgery, and in the identification of mandibular canal involvement in dentoalveolar surgery.2Corbella S. Srinivas S. Cabitza F. Applications of deep learning in dentistry.Oral Surg Oral Med Oral Pathol Oral Radiol. 2020; S2212-4403: 31321Google Scholar Although these systems may theoretically reduce variation in clinical practice and prevent medical error, their current limitations must also be understood. To this end, oral-maxillofacial surgeons (OMSs) are at a crossroads: should OMSs begin to adopt AI usage? To do so, education involving AI and improvements in publication quality will be expected. In preparation for the implementation of AI, oral-maxillofacial surgery residents and attendings must optimize training programs, understand its clinical applicability, and initiate a conversation on how it may be utilized and regulated in routine clinical care. Although it may not be necessary for all OMSs to achieve a full understanding of the mathematical structure of AI algorithms, a basic discussion on the workings of these models is required. AI uses deep learning algorithms with convolutional neural networks, which are based on networks that replicate the mechanism of human neurons. These models have replaced traditional methodology in detection and diagnostic systems by allowing extrapolation from associated databases to be actively learned over time.3Kuwada C. Ariji Y. Fukuda M. et al.Deep learning systems for detecting and classifying the presence of impacted supernumerary teeth in the maxillary incisor region on panoramic radiographs.Oral Surg Oral Med Oral Pathol Oral Radiol. 2020; 130: 464Abstract Full Text Full Text PDF PubMed Scopus (30) Google Scholar In essence, these systems learn as humans do. The more educated the system is with high-quality data, the more accurate it becomes in its abilities. There may be benefits to the implementation of an automated diagnostic and prognostic system that precisely identifies pathologies, fractures, and diseases, and provides prognoses with ideal treatment options. However, before we may begin to comprehend and apply AI to these real-world clinical situations, the conversation may best be framed by first discussing its current limitations. First, AI systems are limited by an inability to explain its decision-making in an understandable manner. Essentially, these models are considered black-boxes and as they increase in performance, they become less explainable. Currently, it is impossible to dissect and understand their workings on a computational level.1Kelly C.J. Karthikesalingam A. Suleyman M. et al.Key challenges for delivering clinical impact with artificial intelligence.BMC Med. 2019; 17: 195Crossref PubMed Scopus (460) Google Scholar While transparency may not be necessary on a research or theoretical basis, transparency in health care is critical. Patients and providers may find it difficult to trust an algorithm that may provide accurate, yet unexplainable diagnoses and prognoses. The development of explainable AI models and informed clinicians may facilitate their adoption and improve their level of transparency in treatment. Most AI models use historically labeled, retrospective data to train and implement algorithms. As models encounter new data that differs from data encountered in the training of a retrospective-based model, performance may be suboptimal. Furthermore, most AI studies are not published in reputable peer-reviewed journals.1Kelly C.J. Karthikesalingam A. Suleyman M. et al.Key challenges for delivering clinical impact with artificial intelligence.BMC Med. 2019; 17: 195Crossref PubMed Scopus (460) Google Scholar Once these models are built on stable, prospective databases, and reviewed and published in high-quality journals, the true ability of AI systems may become apparent, leading to application in clinical practice. A final limitation is the “AI chasm”: reported model accuracy may not represent clinical efficacy.1Kelly C.J. Karthikesalingam A. Suleyman M. et al.Key challenges for delivering clinical impact with artificial intelligence.BMC Med. 2019; 17: 195Crossref PubMed Scopus (460) Google Scholar A common, utilized metric in AI studies is the area under the curve of a receiver operating characteristic curve. This metric may not be the ideal measurement of clinical applicability and its comprehension is complicated for those with limited experience in AI. AI studies must publish information regarding positive and negative predictive values that are relevant to changes in patient outcomes. It will be sensible of oral-maxillofacial surgery to develop an understanding of basic AI end points and metrics. More so, AI models used in patient care must be centered around patient outcomes and clinical applicability, instead of theoretical model accuracy. In conjunction with the understanding of AI model limitations, OMSs must also closely monitor its regulation. The U.S. Food and Drug Administration started developing a framework to ensure safe and effective AI models can progress to patient care.4Artificial intelligence and Machine learning in Software as a medical Device.https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-deviceDate accessed: February 5, 2021Google Scholar While this is promising, it is also inherently a challenge. Given the intrinsic nature of AI learning, systems may improve and change in performance in time, prompting a more advanced evaluation process. Ongoing performance monitoring guidelines must be developed to continually calibrate AI models. The OMS should be familiar with any regulatory progress regarding the usage of AI in patient care. As AI is developing, OMSs should begin to adapt to its eventual use. To improve the understanding of AI models and algorithms, oral-maxillofacial surgery residencies should incorporate basic AI curriculums into their programs. Program directors may invite experts in healthcare AI to teach on a foundational, clinical level. This exposure may allow residents to begin to critically appraise models for their use and enable residents to adopt AI tools safely in their practice. Simultaneously, to optimize future AI models, oral-maxillofacial surgery residents should be encouraged to publish high-quality, prospective studies throughout their training. Once AI models are accepted into routine care, basing AI models on more predictable, prospective databases will provide the OMS a seamless transition to AI usage. OMSs should encourage the collaboration between groups around the world to establish high-quality AI training data covering the entire of spectrum of patients. Such actions may ensure that OMSs develop a significant role in the usage of AI in the maxillofacial region and continue to innovate the field. The applicability of AI in oral-maxillofacial surgery is limitless. Its implementation into diagnostic processing may act as an aid in identifying pathologies and fractures. Models built on high-quality databases may provide important prognostic indicators that may guide surgeries and postoperative management. For example, developing a model that may compute the likelihood of implant failure by assessing imaging and incorporating medical comorbidities, drugs, and other potential risk factors may revolutionize implantology. Developing the first clinical model in this area may strengthen OMSs' role in dental implants. A plethora of dental imaging is also available to the OMS. With its use, oral-maxillofacial surgery may use AI to screen for various pathologies present in panoramic imaging and the surrounding head and neck region to increase the role oral-maxillofacial surgery plays in the medical management of patients. However, these innovations will not be possible unless OMSs adequately prepare for AIs arrival in clinical practice. By understanding its limitations and regulatory progress, and implementing AI education training into residency programs, OMSs may become the leading figures involving AI in maxillofacial diagnosis and prognosis and push the field into this new technological era.

Full Text

Published Version
Open DOI Link

Get access to 115M+ research papers

Discover from 40M+ Open access, 2M+ Pre-prints, 9.5M Topics and 32K+ Journals.

Sign Up Now! It's FREE

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call