歯科パノラマ画像からのディープラーニングと 最適化法を用いた歯列認識

抄録

Enormous number of panoramic X-ray images are checked by dentists every day. It needs much time for dentists and may cause interpretation errors. There are some studies on teeth detection with deep learning. However, object detection using deep learning provides multiple candidates, and it is not east to select one from the candidates. This paper proposes an automatic teeth recognition method, which refines the candidates by an optimization method with prior knowledge model. At first the proposed method detects teeth candidates by faster R-CNN, which is one of the deep learning-based techniques. Next, it determines the best candidate by optimizing an objective function. The objective function evaluates the relative position of the teeth based on prior knowledge model. With 1000 images, the accuracy of the proposed method was 0.94. It is higher than 0.92 that is without optimization. The proposed method performs the better recognition accuracy in comparison with the method without optimization step.