Synchronization in retrospective respiratory gating using Deep Learning

Breathing motion introduces artifacts during CT acquisition, what affect the quality and subsequent reconstruction of the images. This study aims to reduce artefact in CT images using deep learning techniques. Specifically, we propose the implementation of an autoencoder based on convolutional neural networks. Once the model was trained, we employed a morphing technique to generate new images with reduced respiratory motion. By analyzing the respiratory signal, we classified the different images into phases and selected those most suitable for correction. Subsequently, we applied the de- scribed method, obtaining a more homogeneous data set. The results demonstrate a significant reduction in motion when comparing intensity changes within the regions most affected by motion. Thus, we validated the efficacy of the proposed approach to mitigate breathing-induced artifacts. The appli- cation of artificial intelligence (AI) in this field represents a significant advance. This study provides promising initial results and opens up new possibilities for research and de- velopment. By complementing existing techniques, AI offers enhanced motion reduction capabilities, thus improving the quality of CT images. The potential for future advances in this direction are substantial, promising further reduction in respiratory motion artifacts and improvement to the overall quality of CT images.