Masume Johari1, Farzad Esmaeili1, Alireza Andalib2, Shabnam Garjani1, Hamidreza Saberkari3. 1. 1 Department of Oral and Maxillofacial Radiology, Faculty of Dentistry, Dental and Periodontal Research Center, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 2. 2 Department of Electrical Engineering, Tabriz Branch, Islamic Azad University, Tabriz, Islamic Republic of Iran. 3. 3 Department of Electrical Engineering, Sahand University of Technology, Tabriz, Iran.
Abstract
OBJECTIVES: Detection of vertical root fractures (VRFs) in their initial stages is a crucial issue, which prevents the propagation of injury to the adjacent supporting structures. Designing a suitable neural network-based model could be a useful method to diagnose the VRFs. The aim of this study was to design a probabilistic neural network (PNN) to diagnose the VRFs in intact and endodontically treated teeth of periapical and CBCT radiographs. Also, we have compared the efficacy of these two imaging techniques in the detection of VRFs. METHODS: A total of 240 radiographs of teeth (120 radiographs of teeth with no VRFs and 120 teeth with vertical fractures, with half of the teeth in each category treated endodontically and the remaining half intact, i.e. not endodontically treated) were used in 3 groups for training and testing of the neural network as follows: Group 1, 180/60; Group 2, 120/120; and Group 3, 60/180. First, Daubechies 3 wavelet was applied to acquire the image analysis coefficients on two planes; then Gabor filters were used to extract the image characteristics, which were used to educate the PNN. The designed neural network was able to diagnose and classify teeth with and without VRFs. In addition, in order to determine the best training and test sets in the network, the variance of the function of network changes was manipulated at a range of 0-1 and the results were assessed in terms of the parameters evaluated, including sensitivity, specificity and accuracy. RESULTS: In the periapical radiographs, the maximum accuracy, sensitivity and specificity values in the three groups were 70.00, 97.78 and 67.7%, respectively. These values in the CBCT images were 96.6, 93.3 and 100%, respectively, at the variance change range of 0.1-0.65. CONCLUSIONS: The designed neural network can be used as a proper model for the diagnosis of VRFs on CBCT images of endodontically treated and intact teeth; in this context, CBCT images are more effective than similar periapical radiographs. Limitations of this study are the use of sound one-rooted premolar teeth without carious lesions and dental fillings and not simulating the adjacent anatomic structures. Further in vitro work using a full-skull simulation for CBCT and skin/bone simulation is needed.
OBJECTIVES: Detection of vertical root fractures (VRFs) in their initial stages is a crucial issue, which prevents the propagation of injury to the adjacent supporting structures. Designing a suitable neural network-based model could be a useful method to diagnose the VRFs. The aim of this study was to design a probabilistic neural network (PNN) to diagnose the VRFs in intact and endodontically treated teeth of periapical and CBCT radiographs. Also, we have compared the efficacy of these two imaging techniques in the detection of VRFs. METHODS: A total of 240 radiographs of teeth (120 radiographs of teeth with no VRFs and 120 teeth with vertical fractures, with half of the teeth in each category treated endodontically and the remaining half intact, i.e. not endodontically treated) were used in 3 groups for training and testing of the neural network as follows: Group 1, 180/60; Group 2, 120/120; and Group 3, 60/180. First, Daubechies 3 wavelet was applied to acquire the image analysis coefficients on two planes; then Gabor filters were used to extract the image characteristics, which were used to educate the PNN. The designed neural network was able to diagnose and classify teeth with and without VRFs. In addition, in order to determine the best training and test sets in the network, the variance of the function of network changes was manipulated at a range of 0-1 and the results were assessed in terms of the parameters evaluated, including sensitivity, specificity and accuracy. RESULTS: In the periapical radiographs, the maximum accuracy, sensitivity and specificity values in the three groups were 70.00, 97.78 and 67.7%, respectively. These values in the CBCT images were 96.6, 93.3 and 100%, respectively, at the variance change range of 0.1-0.65. CONCLUSIONS: The designed neural network can be used as a proper model for the diagnosis of VRFs on CBCT images of endodontically treated and intact teeth; in this context, CBCT images are more effective than similar periapical radiographs. Limitations of this study are the use of sound one-rooted premolar teeth without carious lesions and dental fillings and not simulating the adjacent anatomic structures. Further in vitro work using a full-skull simulation for CBCT and skin/bone simulation is needed.
Authors: Igor Tsesis; Kivanc Kamburoğlu; Alexander Katz; Aviad Tamse; Israel Kaffe; Anda Kfir Journal: Oral Surg Oral Med Oral Pathol Oral Radiol Endod Date: 2008-03-04