Meeting Banner
Abstract #2295

Combining Convolutional Neural Networks, Multiparametric MRI, and Error Detection to Improve Automated Liver Segmentation

Matthew Gibbons1, Edgar Castellanos Diaz1, Suneil K Koliwad2, Peter W Hunt2, Jean-Marc Schwarz2,3, Kathleen Mulligan2,3, Robert H Lustig4, Alejandro Gugliucci3, Diana L Alba2, Ayca Erkin-Cakmak4, and Susan M Noworolski1
1Department of Radiology and Biomedical Imaging, University of California San Francisco, San Francisco, CA, United States, 2Department of Medicine, University of California San Francisco, San Francisco, CA, United States, 3College of Osteopathic Medicine, Touro University - California, Vallejo, CA, United States, 4Department of Pediatrics, University of California San Francisco, San Francisco, CA, United States

Synopsis

The objective of this study was to generate an automatic liver segmentation method. Two methods were compared. The first, M1, was a Convolutional Neural Network (CNN) trained on proton density fat fraction (PDFF) maps. The second, M2, was the CNN trained on multiparametric MRI (mpMRI) images combined with an error detection protocol. The distributions for Dice similarity coefficient (DSC), volume, and PDFF were improved for M2 versus M1. The DSC mean increased from 0.91 to 0.96. The M2 method was effective in detecting and correcting poor segmentations while significantly reducing processing time as compared to manual segmentation.

This abstract and the presentation materials are available to members only; a login is required.

Join Here