Skip to Main content Skip to Navigation
Master Thesis

Segmentation de lésions carieuses à partir de radiographies par un réseau de neurones à convolution

Abstract : The early detection of caries lesions is an essential aspect in the treatment of caries disease in order to be able to intervene as quickly as possible and preserve dental tissue. The diagnosis of caries disease is based on physical examination, but lesions may be detected on X-ray examination. In recent years, the field of artificial intelligence has expanded with the development of convolutional neural networks, some of which specialized in the detection and segmentation of medical images. The aim of this thesis was to program a convolutional neural network able to detect and segment caries lesions from an intraoral X-ray. One thousand three hundred and seventy-five radiographs were analyzed and the carious lesions segmented. These data were used to train the network to segment lesions and test its performance. The results obtained show that it is possible to obtain a functional network with actionable results. However, the values of sensitivity and specificity measured reflect too many errors of underdiagnosis and therefore an inability to use it clinically. These results nevertheless demonstrated that convolutional neural networks are functional, applicable in the dental field and it appears necessary to continue research in order to optimize the results. The literature currently agrees that artificial intelligence cannot replace humans in the diagnosis of caries lesions.
Complete list of metadata
Contributor : Odonto UB - Connect in order to contact the contributor
Submitted on : Friday, November 26, 2021 - 3:43:12 PM
Last modification on : Friday, December 17, 2021 - 3:47:07 AM
Long-term archiving on: : Sunday, February 27, 2022 - 7:42:21 PM


Files produced by the author(s)


  • HAL Id : dumas-03451688, version 1



Van-Hoan Nguyen. Segmentation de lésions carieuses à partir de radiographies par un réseau de neurones à convolution. Sciences du Vivant [q-bio]. 2021. ⟨dumas-03451688⟩



Record views


Files downloads