Sagreiya H., Akhbardeh A., Durot I., Rubin D.L.
University of Pennsylvania, US
Keywords: hepatic fat, informatics, machine learning, MRI, ultrasound
Nonalcoholic fatty liver disease is a global pandemic. This study investigated whether ultrasound point shear wave elastography measurements could predict hepatic fat quantification using a machine larning (ML) algorithm trained with fat quantification on MRI. 186 exams from Stanford and 50 from the University of Wisconsin were analyzed. Hepatic fat values were quantized into intervals of 5%, and a multi-model support vector machine (SVM) was run with 10 measurements of shear wave velocity as inputs. For each fat quantification level, a dedicated SVM was trained; the overall fat prediction was determined by fusing model results. Validation was via leave-one-out cross validation. Pearson correlation was calculated between predicted and actual fat quantification. There was a high correlation between ML-predicted fat quantification and MR-based fat quantification for Stanford (r=0.98) and Wisconsin (r=0.95) data. ML correctly predicted the fat quantification interval for most subjects.
Journal: TechConnect Briefs
Volume: TechConnect Briefs 2021
Published: October 18, 2021
Pages: 105 - 108
Industry sectors: Advanced Materials & Manufacturing | Medical & Biotech
Topics: Advanced Manufacturing, Materials Characterization & Imaging
ISBN: 978-0-578-99550-2