Automatic detection of cephalometric points using deep learning

Today, large amounts of data are collected using computers in every sector. Using machine learning methods in areas such as networking, defense industry, space and cybersecurity, these data can be reported with high success rates and meaningful information can be extracted. With the increase in research in digital x-ray imaging, experts have successfully brought forward interesting and effective methods to address critical medical analysis problems. One of these fields is cephalometric analysis. Within the scope of the study of this thesis, a solution was presented with deep learning on cephalometric image analysis. Cephalometric analysis is used for disease diagnoses, evaluation in oral and maxillofacial surgery areas and craniofacial growth estimate, treatment plan, curative effect evaluation and comparison of different cases. In this thesis study, a U-Net model was developed that makes automatic detection of Cephalometric points using Convolutional Neural Networks. The data used in the scope of this research are obtained from the publicly available dataset provided during the 2015 ISBI Grand Challenge Training Dataset. 19 Cephalometric points are detected automatically. 74,0% Success Detection Rate was achieved in the range of 2 mm, 81,4% in the 2.5 mm range, 86,3% in the 3mm range and 92,2% in the 4mm range.