Generalizability of Deep Learning Models for Caries Detection in Near-Infrared Light Transillumination Images

Abstract

Objectives: The present study aimed to train deep convolutional neural networks (CNNs) to detect caries lesions on Near-Infrared Light Transillumination (NILT) imagery obtained either in vitro or in vivo and to assess the models’ generalizability. Methods: In vitro, 226 extracted posterior permanent human teeth were mounted in a diagnostic model in a dummy head. Then, NILT images were generated (DIAGNOcam, KaVo, Biberach), and images were segmented tooth-wise. In vivo, 1319 teeth from 56 patients were obtained and segmented similarly. Proximal caries lesions were annotated pixel-wise by three experienced dentists, reviewed by a fourth dentist, and then transformed into binary labels. We trained ResNet classification models on both in vivo and in vitro datasets and used 10-fold cross-validation for estimating the performance and generalizability of the models. We used GradCAM to increase explainability. Results: The tooth-level prevalence of caries lesions was 41% in vitro and 49% in vivo, respectively. Models trained and tested on in vivo data performed significantly better (mean ± SD accuracy: 0.78 ± 0.04) than those trained and tested on in vitro data (accuracy: 0.64 ± 0.15; p < 0.05). When tested in vitro, the models trained in vivo showed significantly lower accuracy (0.70 ± 0.01; p < 0.01). Similarly, when tested in vivo, models trained in vitro showed significantly lower accuracy (0.61 ± 0.04; p < 0.05). In both cases, this was due to decreases in sensitivity (by −27% for models trained in vivo and −10% for models trained in vitro). Conclusions: Using in vitro setups for generating NILT imagery and training CNNs comes with low accuracy and generalizability. Clinical significance: Studies employing in vitro imagery for developing deep learning models should be critically appraised for their generalizability. Applicable deep learning models for assessing NILT imagery should be trained on in vivo data.

Deep learning for caries detection: A systematic review

“…Fifth, we found a wide range of imagery to be employed for deep learning tasks. Some, like near-infrared light transillumination or optical coherence tomography, are clinically uncommon, and interpretation of such images may not be accurate in the hands of inexperienced examiners [18,[39][40][41]. The latter two types of imagery, however, are clinically promising, as they show high accuracy and do not require ionizing radiation, which is why deep learning seems useful to employ here overcoming the described "experience gap".…”

Analysis of Deep Learning Techniques for Dental Informatics: A Systematic Literature Review

“…The proposed MI-DCNNE technique was more successfully implemented by using a score-based ensemble approach with 99.13% accuracy score [ 87 ]. Several researchers demonstrate a DL strategy for identifying and localizing dental lesions [ 88 ] in TI images automatically and dental carries in NILT images [ 89 , 90 ] and on the children’s first permanent molar [ 91 ]. Their research shows that using a DL technique to analyze dental photos can improve caries detection speed and accuracy, as well as complement dental practitioners’ diagnosis and improve patient outcomes [ 88 ].…”

“…Different types of images were used by different researchers based on the techniques they used in DI. Radiographic images [ 16 , 41 , 43 , 56 , 63 , 64 , 65 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 87 , 157 ], near-infrared light transillumination (NILT) [ 88 , 89 , 90 ], intraoral images [ 66 , 86 , 91 , 92 , 93 , 95 , 96 , 97 , 158 , 159 , 160 ], 3D model [ 102 , 113 , 114 , 115 , 161 ] were used in the research for dental diseases diagnostic on the 3D dental model. The studies on the dental disease’s diagnostic on the CBCT dental model used the CT images [ 124 ], 3D CT scans […”

A Comprehensive Review of Recent Advances in Artificial Intelligence for Dentistry E-Health

“…Tables 12 and 13 provide a risk assessment summary for each study, indicating the level of bias based on predefined criteria for both X-ray-based and NILT-based imagery. Machine learning Spectral image enhancement for dental disease diagnosis (low) [131] Disease detection Machine learning Dental caries detection using NIR imaging (low) [132][133][134] Disease classification Classical image analysis approaches Dental tissue classification using NIR hyperspectral imaging (low) [135,136] Deep learning Dental caries classification using CNNs (moderate) [137] Disease segmentation Deep learning Lesion segmentation using deep CNN (moderate) [138,139]…”