Akademik Veri Yönetim Sistemi
The 23rd International Congress of DentoMaxilloFacial Radiology, Seoul, Güney Kore, 28 Nisan - 01 Mayıs 2021, ss.37, (Özet Bildiri)
Introduction
This study aims to test the potential of the anatomical landmarks designated for a research related to hand-wrist maturation and ma- chine learning, by preparingsubset of sample with the envisaged points to develop temporary models and evaluate the results of the classification. Thus, the potential of the envisaged anatomical land- marks is evaluated with a pilot study to prevent any possible loss of time.
Materials and Methods
This study was conducted with 136 hand-wrist radiographs taken at the Medipol Mega University Hospital, Faculty of Dentistry, De- partment of Orthodontics. Therelationship of the epiphysis and diaphysis were categorized as ‘narrow’, ‘equal’, ’capping’ or ‘fusion’.
A total of 22 anatomical points is designated to be used in extracting the features of the DP3, MP3, PP3 and MP5 regions. A web-based system is developed forlabeling of the data set. The samples are uploaded and labeled by three observers. 18 ratios and 15 angles were extracted and used to develop two different neuralnetworks, NN-1 and NN-2. The percentage of agreement, Cohen’s Kappa (cκ) and Weighted Kappa (wκ) coefficients are calculated to evaluate the results of the models.
Results
The cκ coefficients for the method error were in between 0.7 and 1. The compliance of the classification results and the data labels were changed between82.14% and 89.29%. On average, NN-1 and NN-2 models were shown 85.71% and 85.52% performance in classifying the relationship of the epiphysis andthe diaphysis. The cκ and wκ coefficients of the NN-1 model for each phalanx region changed be- tween -0.01 and 0.91.
Conclusions
The anatomical points envisaged to be used in determining the pha- lanx morphology are found to be promising.