Abstract

This is the augmentation dataset used in the paper named "LSTformer: Long Short-term Transformer for Real Time Respiratory
Prediction". We made an augmentation dataset utilizing an RGB-D camera to collect motion signals in a breathing simulator phantom device. It is worth noticing that the movement of the simulator is driven by the clinical patient’s respiration, which is from a public dataset (https://signals.rob.uni-luebeck.de/index.php). Other details can be seen in our previous work : H. Peng, L. Deng, Z. Xia, Y. Xie, and J. Xiong, “Unmarked external breathing motion tracking based on b-spline elastic registration,” in International Conference on Intelligent Robotics and Applications. Springer, 2021, pp. 71–81.

Instructions:

To increase the diversity of the public dataset and improve the generalizability of the trained model, we obtained the augmented dataset via a breathing simulator. we trained our LSTformer model by randomly picking 97 training data segments (90% dataset). The rest of the 9 data segments (10% dataset) are for testing, and each segment contains 7500 points of respiratory motion.