AFFINITY FOR VICINITY: TRANSFORMER BASED CEPHALOMETRIC LANDMARK DETECTION

Received: 19.06.2021; Revised: 28.07.2021, Accepted: 21.09.2021, Published Online: 26.10.2021

Dr. Agam Das Goswami

School of Electronics Engineering, VIT-AP University, India, mailtoap2021@gmail.com

Pravan Pradeep

School of Electronics Engineering, VIT-AP University, India

Tathagat Banerjee

School of Electronics Engineering, VIT-AP University, India

Abstract— In orthodontic treatment the automation of Cephalometric landmarks mark an essential role. The idea is to train an artificial intelligent software such that it can predict the landmarks with as high mean average precision as possible. The development of a robust technique shall not only help clinical detection but also aid the upcoming planned different architectures and research. The paper proposes the use of state-of-the-art transformers to synthesise an attention based model to detect landmarks. It acknowledges the prediction tendency a human uses for annotation, historical training, last marked output(localisation) and generic outputs(globalisation). We have incorporated the help of 1100 image samples from different medical regions. We utilise the idea of Intersection of union and loss metrics to train our model. This paper proposes the first architecture of symbiosis of transformer and cephalometric landmark detection.

Keywords— Cephalometric landmark detection, Biomedical imaging, Transformer, Artificial intelligence, Attention, Feature matrix.