Impact of data labeling protocol on the quality of LGE-MRI atrial segmentation

Abstract

AbstractAtrial fibrillation affects up to 2% of the adult population in developed countries, and ablation as the main method of treatment leads to a high probability of recurrence. For such procedures, the approach of creating an in silico model of the patient’s atrium to be used for navigation during the catheter ablation procedure itself is extremely promising. In this case, the MRI data on which the model is based must be loaded into the system and segmented with high accuracy. This paper describes a new universal protocol for the segmentation of LGE MRI images. This protocol has been used to train state-of-the-art neural networks for automatic MRI segmentation. It is shown that the new data labeling protocol significantly improves the training quality of the network. Using this approach, it is possible to improve the quality of the reproduction of the patient’s atrial parameters and the performance of all related services. The presented protocol is also accompanied by a labeled image dataset. In the future, the data from such labels can be used for predictive modeling and the creation of digital twins of patients’ atria.