Attention Mechanism Enhances Clinical Interpretability of Convolutional Neural Network for Differentiating Benign from Malignant Focal Liver Lesions on Contrast-Enhanced Ultrasound
ePoster
Authors: Thodsawit Tiyarattanachai, Stanford University Simona Turco, Eindhoven Univeristy of Technology John Eisenbrey, Thomas Jefferson University Corinne Wessner, Thomas Jefferson University Alexandra Medellin, University of Calgary Stephanie Wilson, University of Calgary7 Andrej Lyshchik, TJUH Aya Kamaya, Stanford Ahmed Kaffas, Stanford ,
In this work, we incorporated Attention mechanism into convolutional neural network (CNN) for differentiating benign from malignant focal liver lesions on contrast-enhanced ultrasound (CEUS). The Attention mechanism enhanced model interpretation. Analysis of Attention weights revealed the timepoints of CEUS that the model paid attention to, which correlated with clinical interpretation of focal liver lesions on CEUS.