Table 1 Dice’s coefficients of nnU-Net and comDL segmentation for cine and real-time CMR.

From: Assessment of deep learning segmentation for real-time free-breathing cardiac magnetic resonance imaging at rest and under exercise stress

n = 15

LV

MYO

RV

nnU-Net cine ED

0.98 (0.00)

0.91 (0.02)

0.92 (0.05)

nnU-Net cine ES

0.92 (0.04)

0.91 (0.03)

0.88 (0.06)

nnU-Net cine

0.95 (0.02)

0.91 (0.02)

0.90 (0.03)

nnU-Net RT

0.94 (0.02)

0.89 (0.02)

0.90 (0.03)

nnU-Net RT stress

0.92 (0.03)

0.85 (0.03)

0.83 (0.11)

nnU-Net RT max stress (n = 12)

0.91 (0.03)

0.83 (0.04)

0.79 (0.16)

comDL cine ED

0.98 (0.02)

0.97 (0.02)

0.92 (0.06)

comDL cine ES

0.95 (0.05)

0.95 (0.04)

0.88 (0.08)

comDL cine

0.97 (0.03)

0.96 (0.02)

0.90 (0.06)

comDL RT

0.93 (0.04)

0.88 (0.05)

0.92 (0.05)

comDL RT stress

0.79 (0.15)

0.72 (0.15)

0.79 (0.14)

comDL RT max stress (n = 12)

0.70 (0.21)

0.62 (0.19)

0.69 (0.18)

0.94 (0.04)

0.88 (0.02)

0.87 (0.06)

Inter-observer cine (n = 50)15

0.92 (0.04)

0.87 (0.03)

0.88 (0.05)

0.93 (0.04)

0.88 (0.02)

0.89 (0.05)

  1. The table features the mean and standard deviation (in parenthesis) of the Dice’s coefficients for the left ventricular endocard (LV), the left ventricular myocardium (MYO), and the right ventricle (RV) for all volunteers. For cine CMR, images were separated into end-diastolic (ED) and end-systolic (ES) phase. For RT max stress, only data of 12 volunteers were analyzed, because in three cases the image quality was too poor to create reasonable reference contours. Mean and standard deviation (in parenthesis) of previously reported values for inter-observer variability from three different human observers for cine CMR are shown for comparison.
Back to article page