Automated liver segmentation across different imaging modalities. Segmentation volumes and liver biometry can be derived using this segmentation algorithm.
Lobar liver segmentation is a core functionality required to automate liver biometry (measurement of liver volume/mass, fat content, etc) and to delineate segmental location of liver masses. It is also required for surgical and interventional procedure planning, such as pre-operative interventional bland embolization or dose calculation for Y-90 radioembolization.
CT, T1 & T2-weighted MRI
Model performance not tested across
other scanner manufacturers (other than GE and Siemens), imaging phases, and institutional
Information on training data
Trained with unenhanced low-flip-angle two-dimensional (2D) multiecho spoiled gradient-echo(SPGR) MR images with variable T2* weighting (n = 300) with multiple echo times (TEs) to be robust against different signal weightings.
Used transfer learning to generalize CNN to other imaging modalities by using multimodal image data (30 2D SPGR MRI datasets, 10 contrast-enhanced CT datasets, 20 contrast-enhanced T1-weighted hepatobiliary
phase MRI datasets).
Model performance metrics
For the multimodal CNN, Dice scores were 0.92 ± 0.05 for 2D
SPGR (first echo), 0.93 ± 0.02 for 3D SPGR (first echo), 0.94
± 0.06 for contrast-enhanced and unenhanced CT, and 0.95
± 0.03 for HBP T1-weighted MRI.
Agreement of liver volume assessments between convolutional neural network (CNN)−predicted and manual liver segmentation. Linear regression (left) and Bland-Altman analysis (right) of liver volume estimates from contrast-enhanced hepatobiliary phase T1-weighted MRI (HBP-T1w-MRI).
See reference publication for more information.
Model available on Arterys.