Dentistry: The Use of Artifical Intelligence

artificial intelligence

Abstract

Background/purpose

In the past few years, artificial intelligence (AI) has penetrated into the dental field. The purpose of this systematic review is to determine the development of AI applications that are widely used in dentistry, and to evaluate their performance in terms of diagnosis, clinical decision-making and predicting treatment prognosis.

Materials and methods

Through a comprehensive search in electronic databases published in the past two decades (January 2000) (such as PubMed, Medline, Embase, Cochrane, Google Scholar, Scopus, Web of Science and Saudi Digital Library), this article was identified and selected Literature. – March 15, 2020). After applying the inclusion and exclusion criteria, 43 articles were completely read and strictly analyzed. Use QUADAS-2 for quality analysis.

The Challenges and Ways Forward

Despite the unlimited potential, AI solutions have not yet been widely used in conventional medical practice. For example, since 2015, in the dental field, convolutional neural networks have only been used in research environments, mainly on dental X-ray films, and the first applications involving these technologies have now entered the clinical field (Schwendicke et al., 2019).

This is even more surprising when acknowledging that dentistry is particularly suitable for AI tasks:

1) In the dental field, from screening to treatment planning and implementation, images play an important role and cornerstone in most patients’ dental travel.

2) The dentistry regularly uses different imaging materials from the same anatomical area of ​​the same person, and regularly accompanied by non-imaging data, such as clinical records and general and dental history data, including systemic conditions and medications. In addition, data is usually collected at multiple points in time. AI is suitable for effectively integrating and cross-linking these data and improving diagnosis, prediction and decision making.

3) Many dental diseases (caries, apical lesions, periodontal bone loss) are relatively common. A data set with a large number of “affected” cases can be managed with limited work.

We have seen three main reasons why dentistry has not fully adopted AI technology. Addressing these reasons will help improve dental AI technology and promote its application in clinical care.

First, due to data protection issues and organizational barriers, medical and dental data is not as easy to obtain and access as other data. Data is usually locked in a separate, personalized system with limited interoperability. At least compared to other datasets in the AI ​​field, datasets lack structure and are usually relatively small. Each patient’s data is complex, multidimensional and sensitive, and the options for triangulating or validating them are very limited. For example, medical and dental data from electronic medical records show low variable integrity, where data is often systematically lost rather than random. Sampling usually leads to selection biases, either being sick (such as hospital data), or overly healthy (such as data collected by wearable devices), or overly wealthy (such as data on dental treatment in countries lacking universal health insurance) excessive.

Second, in dental AI research, processing data and measurement and verification results are usually not sufficiently replicated and enhanced (Schwendicke et al. 2019). It is not clear how to select, organize and preprocess the data set. Data is often used for training and testing, leading to “data monitoring bias” (Gianfrancesco et al., 2018; England and Cheng, 2019). It is often impossible to define a “hard” gold standard, and there is no consensus on how many experts are needed to mark data points and how to merge the different labels of this “fuzzy” gold standard (Walsh 2018).

Third, the results of dental AI are usually not easy to apply: the single information provided by most dental AI applications today can only partially provide information for the required and complex decisions in clinical care (Maddox et al. 2019). In addition, issues regarding accountability and transparency still exist.

Written by 

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.