Effects of Visual Context on Information Processing during Tooth Identification

Iwahashi, Ryo

doi:10.5466/ijoms.19.295

Cited by 2 publications

(1 citation statement)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In some static images for the training data, the neighboring unneeded teeth were included, and on the other hand, some parts of the target teeth were outside the image, which might impair the training effect. In humans, the cognitive information process differs between the identification of a single tooth and that of multiple teeth in the dentition (31) , and also between the identification of complete and fragmented teeth (30) .…”

Section: Table 3．probability Ofmentioning

confidence: 99%

Application of a Deep Learning Artificial Intelligence System for Individual Tooth Identification

et al. 2021

View full text Add to dashboard Cite

The present study aimed to construct an efficient deep learning (DL)based artificial intelligence (AI) system for identifying isolated teeth for practical use in the fields of forensic dentistry and osteoarchaeology. We first constructed a learning model to identify mandibular incisors and first molars to verify whether the DLbased system could identify class traits, followed by other learning models to identify mandibular premolars to verify whether the DLbased system could identify type traits with large individual variation. The study materials used were dental plaster casts of 16 dental students. The buccal (labial) , lingual, and occlusal surfaces of the cast teeth were filmed using a video camera, and static images of the mandibular central incisor, first molar, and premolars were made and used as the training, validation, and test data for DL. We used a convolutional neural network (CNN) as the DL model, AlexNet as the CNN architecture, convolutional architecture for fast feature embedding (Caffe) as the DL framework, and the stochastic gradient descent (SGD) as the solver type. The results indicated that our learning model could identify incisors and first molars perfectly, but not premolars.These findings suggest that identifying tooth types with large individual variation remains a difficult task, even for AI. Therefore, the numbers and kinds of training data need to be improved to increase the concordance rates for identification by DLbased models.

show abstract

Section: Table 3．probability Ofmentioning

confidence: 99%