Automatized Detection and Categorization of Fissure Sealants from Intraoral Digital Photographs Using Artificial Intelligence

Abstract

The aim of the present study was to investigate the diagnostic performance of a trained convolutional neural network (CNN) for detecting and categorizing fissure sealants from intraoral photographs using the expert standard as reference. An image set consisting of 2352 digital photographs from permanent posterior teeth (461 unsealed tooth surfaces/1891 sealed surfaces) was divided into a training set (n = 1881/364/1517) and a test set (n = 471/97/374). All the images were scored according to the following categories: unsealed molar, intact, sufficient and insufficient sealant. Expert diagnoses served as the reference standard for cyclic training and repeated evaluation of the CNN (ResNeXt-101-32x8d), which was trained by using image augmentation and transfer learning. A statistical analysis was performed, including the calculation of contingency tables and areas under the receiver operating characteristic curve (AUC). The results showed that the CNN accurately detected sealants in 98.7% of all the test images, corresponding to an AUC of 0.996. The diagnostic accuracy and AUC were 89.6% and 0.951, respectively, for intact sealant; 83.2% and 0.888, respectively, for sufficient sealant; 92.4 and 0.942, respectively, for insufficient sealant. On the basis of the documented results, it was concluded that good agreement with the reference standard could be achieved for automatized sealant detection by using artificial intelligence methods. Nevertheless, further research is necessary to improve the model performance.

Artificial intelligence-based diagnostics of molar-incisor-hypomineralization (MIH) on intraoral photographs

“…This investigation was reported in accordance with the recommendations of the Standard for Reporting of Diagnostic Accuracy Studies (STARD) steering committee [ 27 ] and recently published recommendations for the reporting of AI studies in dentistry [ 28 ]. The pipeline of methods, mentioned below, was applied and described in previously published reports [ 19 , 20 ].

Fig.

…”

“…Dental photographs were consistently taken with professional single-reflex cameras equipped with a 105-mm macro lens and a macro flash after tooth cleaning and drying [ 19 , 20 ]. All images were stored (jpeg format, RGB colors, aspect ratio of 1:1) and selected for this study project.…”

“…For this purpose, the randomly selected images (batch size = 16) were multiplied by a factor of ~ 5, altered by different transformations (random center and margin cropping by up to 30% each; random deletion removing up to 30%; random affine transformation up to 180°; random perspective transformation up to a distortion of 0.5; and random changes in brightness, contrast, and saturation up to 10%) and resized (300 × 300 pixels). In addition, to compensate for under- and overexposure, all images were normalized [ 19 , 20 ]. Torchvision (version 0.9.1, https://pytorch.org ) in conjunction with the PyTorch library (version 1.8.1, https://pytorch.org ) was used.…”

“…Existing learning results regarding the recognition of basic structures in the existing image set could thus be reused and skipped in the initial training. Training was performed on a university-based computer with the following specifications: RTX A6000 48 GB (Nvidia, Santa Clara, CA, USA); i9 10850 K 10 × 3.60 GHz (Intel Corp., Santa Clara, CA, USA) and 64 GB RAM [ 19 , 20 ].…”

“…Saliency maps were plotted to illustrate image areas that were used by the CNN to make individual decisions. The saliency maps were calculated by back propagating the CNN prediction and visualizing the gradient of the input of the resized images [ 19 , 20 , 32 ].…”