19 September 2022

---

Application of Artificial Intelligence to endodontic microsurgery

Lara Figini

---

In the last 5 years, thanks to advanced surgical technology equipment and materials, the success rate of endodontic microsurgery has greatly improved and still fluctuates between 80% and 94%. Inclusion of teeth with cracks and / or apicomarginal defects more frequently leads to an uncertain prognosis in endodontic microsurgery, therefore, careful preoperative analysis and appropriate case selection contribute to higher success rates. Certain prognostic factors can affect success at a distance therefore it is sometimes problematic to make decisions in the clinical setting when faced with cases in which these types of factors are present. From a clinical perspective, the decision-making process is highly dependent on the experience of the practitioners, so it can lead to bias and unwanted errors. Even for experienced specialists, the analysis of complex cases takes a long time. In recent years, artificial intelligence (AI) has been introduced in the medical area, and with the continued inclusion of new data, AI models have incrementally enhanced the predictive of performance. As a subfield of AI, machine learning (ML) works excellently in predicting the prognosis in dentistry. In the prognosis of dental implantology, for example, with this method, the future average bone levels of the single implant can be predicted based on clinical and radiographic variables. Such tools can simplify treatment options, as well as reduce both unnecessary costs and harms related to unlikely success. Artificial intelligence can also be useful for accurately predicting the results of endodontic microsurgery, but to date, sufficient tools have not yet been developed in this regard.

Materials and methods
 In a very recent study, published in the March 2022 Journal of Dentistry, the authors sought to establish and validate machine learning models for prognosis prediction in endodontic microsurgery, avoiding treatment failure and supporting clinical decision making. A total of 234 teeth from 178 patients were included in this study. Age, sex, tooth type, number of root canals, lesion size, type of bone defect, root filling density, root filling length, and / or apical extension, secondary surgery were considered as prognostic variables and factors. and difficulty of the case. Radiographic measurement was performed using cone beam computed tomography (CBCT) images. Radiographic outcomes were assessed one year after surgery, according to the classification proposed by Rud et Molven. The incomplete healing group, the uncertain healing group, and the unsatisfactory healing group were all classified as a single "unhealed" group. The difficulty of the case was assessed according to pre-established criteria. Cases showing at least one factor listed in the high difficulty category or more than three moderate factors were classified in the difficult group. Cases exhibiting only the low difficulty factors were placed in the least difficulty group. The radiographs were independently reviewed by two experienced operators and any disagreements were resolved by a third endodontic specialist. The gradient increase of the machine model (GBM) and the random forest (RF) have been developed. A 5-fold cross-validation layered approach was used. Predictive Accuracy, Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), F1 Score and Area Under the Curve (AUC), Receiver Operational Characteristics Curve (ROC) ) They were calculated to evaluate predictive performance. 

Results 

Eight important predictors were found, including tooth type, lesion size, type of bone defect, density of root canal filling, length of root filling, and its apical extension, age and sex. For the GBM model, the predictive accuracy was found to be 0.80, the sensitivity of 0.92, the specificity of 0.71, the PPV of 0.71, the NPV of 0.92, the F1 of 0 , 80 and the AUC of 0.88. For the RF model, the precision was found to be 0.80, with a sensitivity of 0.85, specificity of 0.76, PPV of 0.73, NPV of 0.87, F1 of 0.79, and AUC of 0 , 83.

Conclusions

From the data of this study, which must be confirmed in other similar studies, it can be concluded that the trained models developed on the basis of eight common variables, show the potential ability to predict prognosis in endodontic microsurgery. The GBM model offers better guarantees than the RF model with AUC of 0.88 and 0.83 respectively.

Clinical implications

Artificial intelligence (AI) is already able to make diagnosis and prognosis based on a simple radiographic image or on a single photo of a histological preparation. Dentists, and specifically endodontists, can use machine learning models for preoperative analysis in endodontic microsurgery. Models should improve the efficiency of clinical decision making and assist in doctor-patient communication.

For additional information: Machine learning models for prognosis prediction in endodontic microsurgery