Automatic Extraction of Liver Parenchyma and R2* Estimation using Deep Learning for UTE Imaging for Assessment of Hepatic Iron Overload

Ultra-short echo time (UTE) imaging increases the accuracy of R2*-based hepatic iron content (HIC) quantification in cases of high iron overload when conventional GRE sequences can fail due to rapid signal decay. Segmenting whole liver to estimate liver R2* requires human expert and is time consuming. In this study, we trained a convolutional neural network (CNN) to automatically segment the liver parenchyma on radial UTE acquisitions using magnitude images and R2* maps. Our results show an excellent agreement between manual and CNN-based liver segmentation and mean R2* values, hence demonstrating the potential of our proposed method for automated HIC assessment.