Applying Attention Mechanism and Deep Neural Network for Medical Object Segmentation and Classification in X-Ray Fluoroscopy Images.

We study how to apply attention mechanism and deep neural network for real-time segmentation and classification of balloon objects from X-ray fluoroscopy images during percutaneous balloon compression (PBC) surgical procedures. Fast and accurate identification of balloon shape and its relative location to the Meckel’s cave can be of significant benefit to the success of the PBC procedure. In this work, we combine the most successful region-based convolutional neural network pipeline with attention mechanism to address these challenges.