[2Q6-GS-10] Vision, speech: Image analysis and application

[2Q6-GS-10-01] Cephalometric Landmark Location with Multi-phase Deep Learning

Improvement of precision in cephalometric analysis

〇Soh Nishimoto¹ (1. Hyogo College of Medicine)

Keywords:Multi-phase deep learning, Cephalogram, Regression analysis, anatomical landmark

Objective: Cephalometric analysis has long been one of the most helpful approaches in evaluating cranio-maxillo-facial skeletal profile. In analyzing process, locating anatomical landmarks on an X-ray image is a crucial procedure, demanding time and expertise. Development of an automated cephalogram analyzing system will be a great help for practitioners. Deep learning is one of the most developing techniques in these days, in data science field. An automated landmark locating system, utilizing multi-phase deep learning, was developed.

Methods: A regressional system was consisted with convolutional neural networks (CNN) of three phases. With datasets, used in International Symposium on Biomedical Imaging 2015, networks were trained and tested.

Results: The system demonstrated better accuracy than that with single phase CNN. The system presented better results, in comparison with previous benchmarks.