Figure 6:

Ability of active learning–trained 3D U-Net liver volumetry to predict mortality or transplantation compared with standard liver volumetry and Model for End-Stage Liver Disease with sodium (MELD-Na). Split median Mantel-Cox time to death or transplant analysis was performed for predicted average liver volume per ideal body weight, standard volumetry average liver volumes per ideal body weight, and average MELD-Na score in patients where liver volumetry was performed on two or more scans between January 1, 2014, and December 31, 2016. Patients with any history of partial hepatectomy or living donor transplant and those with local-regional intervention after the initial scan were excluded. A, Average predicted liver volume using the Liver Tumor Segmentation plus over- and underestimated active learning cases (LiTS-OU) model per ideal body weight above versus below a median value of 23.9 mL/kg was significantly different (P = .036, χ²= 4.39). B, Radiologist report–extracted average liver volumes per ideal body weight above and below a median value of 20.7 mL/kg was not significantly different (P = .104, χ² = 2.64). C, Average MELD-Na score above and below median value of 8.7 was not significantly different (P = .72, χ² = 0.125).