Artificial Intelligence in Endodontics: A Peek into the Future

Introduction

In recent years, dentistry has witnessed a boom in the technological advancements in the procedures associated with it which has increased the success rates and helped to evaluate the treatment outcomes in a better way. Artificial intelligence (AI) is one such modality that has started to pave its path in the field of endodontics. It has been found useful in the radiodiagnosis of carcinoma and bone age assessment in the medical field.

It is a technology that employs machines and gadgets to simulate human behavior. The virtual component of AI, known as software-type algorithms forms the main module which is used in dentistry.¹

The AI methodologies comprise primarily knowledge-based AI and data-driven AI. The knowledge-based AI is based on a top-down approach to modeling human knowledge and the self-reported knowledge and concepts which humans employ to deduce a solution to a problem. However, it carries the drawbacks of excessive time consumption and preliminary efforts to develop an algorithm based on human knowledge.²

On the other hand, data-driven AI which is commonly known as machine learning (ML), employs a bottomup approach, contrary to knowledge-based AI. In data-driven AI, mathematical models are trained with human activity-derived data. It is further classified into supervised, unsupervised, and semi-supervised learning. ³ The supervised learning module utilizes algorithms like a decision tree and artificial neural networks (ANN) to learn the correlations between data instances and labels. The functioning of ANNs derives its inspiration from the working of the vertebrate nervous system which is a highly interconnected network system and functions similarly by the reception of the signals, followed by mathematical computation and ultimately transmission of the processed data to the next higher level. The unsupervised variant has the potential to analyze data minutely with the use of multilayer convolutional neural networks (CNN).^[1] The semi-supervised learning is the conglomeration of the above two variants. All of these methods ultimately aim to augment clinicians’ competence in dealing with complex and massive data.

Potential applications in Endodontics

Assessment of case difficulty- Predictive models which are based on data-driven AI can be used to diagnose patients vulnerable to tooth structure loss and root caries. A recently published study utilized ANN to predict the level of case difficulty and reported a sensitivity of 94.96% by the ML algorithm.⁴

Determination of root morphology- Deep learning (DL) AI system has been found to detect extra roots in molars on cone-beam computed tomograms (CBCT). Image patches from panoramic radiographs were segmented and augmented and later entered into the DL system to assess their diagnostic performance. Diagnostic accuracy of 86.9% was achieved for the determination of the number of roots present.⁵

Detection of periapical lesions- A study by Orhan et al. aimed to verify the accuracy of the DL system in the detection of a periapical pathosis by an AI system by segmentation of images, followed by volumetric measurement of the pathosis on CBCT images by both manual and AI systems. The authors concluded that AI systems were comparable to the manual segmentation methods and were 92.8% reliable in correctly detecting a periapical lesion.⁶ Another study reported the utility of a DL-based system based on a U-net architecture to detect periapical pathosis. It employed a segmentation method that labeled the voxel of CBCT into five categories such as “lesion” (periapical lesion), “tooth structure,” “bone,” “restorative materials,” and “background.”, following which repeated splits of the images were entered into the DL system to perform cross-validation. The detection accuracy of the DL-based algorithm was found to be 93% with a cumulative DICE index (similarity between two data sets, in this case, the manual method and the DL-based method) of 67% for the true positive lesions.⁷

Cariology- A CNN -based AI system, trained on a semantic segmentation method, was found to generate an area of 83.6% and 85.6% under the receiver-operating characteristic (ROC) curve for occlusal and proximal lesions respectively, signifying an excellent discriminating ability between the presence or absence of carious lesions.⁸

Regenerative Endodontics- When the predictive ability of stem-cell viability under different bacterial lipopolysaccharide concentrations was studied using a neuro-fuzzy system (a form of ANN), a determination coefficient of 0.81 was obtained which shows that 81% of the predictive ability can be accounted for the DL system.⁹

Challenges associated with AI

Despite possessing a lucrative potential, the adoption of AI comes with the following challenges:

i. A colossal amount of data is required for precisive training, which limits its potential to diagnose rare conditions such as periapical lesions other than the endodontic origin. Additionally, healthcare data is not accessible and available readily due to ethical concerns such as maintenance of patients’ privacy, thus rendering only a smaller amount of data patterns for the networks to “learn”.¹⁰ Also, the available data often contains missing information and is also prone to selection bias which leads to the over-representation of a specific data pattern.¹¹

ii. The technology involves complex mechanisms and often it remains uncertain how the datasets are designated, curated, and handled and may persist unvalidated and insufficiently replicable to dental applications.¹⁰

iii. Affordability is also a roadblock to adopting this technology in daily clinical practice as the installation of the machine incurs huge costs.12 Besides, the software programs also require frequent upgradation to adapt to the changing needs.

Conclusion

Experimental research on the applications of AI in the field of endodontics has opened new vistas with its possible applications in the diagnosis, prognosis, and prediction of treatment outcomes. Although deterrents in the form of data acquisition, interpretation, computational power, and moral issues exist, AI can be utilized as an excellent adjunct for dentists by overcoming restraints. Moreover, dental professionals seem to be reluctant to adopt AI in their clinical practice as they are skeptical about the misconception that it can replace clinicians even though it has been well-established that nothing can replace the human brain or intelligence. AI can be utilized as a supplemental tool for the benefit of clinical practice. However, further research is warranted to endorse its usage and potentiate its applications in the realm of endodontics and dentistry at large.

Conflict of interest

None