Fig 1.
T1 VIBE MRI of the neck region for an exemplary volunteer, a) axial view, b) sagittal view, c) coronal view.
Fig 2.
2D US scans of the neck region for an exemplary volunteer.
a) Width and depth of right thyroid lobe marked, b) Length of right thyroid lobe marked.
Fig 3.
Schematic representation of the automatic segmentation workflow.
The raw 3D scans of each lobe are pre-processed by rotation into axial view and resizing as well as centering. Then 2D images are extracted which serve as input to the network. The output of the network is a segmentation from which the volume is calculated by multiplication of pixels which include a segmentation with the pixel volume.
Fig 4.
Illustration of the QuickNAT architecture used here for automatically segmenting thyroid in 3D US volumes.
Fig 5.
Ultrasound image (left), segmentation (middle) and overlay (right) of a good segmentation (first row), a segmentation which segmented too many voxels of the isthmus (second row), a segmentation with too few voxels of the isthmus segmented (third row) and a segmentation with a segmented nodule (fourth row).
Fig 6.
Bland-Altman plots for the intraobserver variability of each medical doctor from Group 2.
All values are given in ml. Row 1, 2 and 3 show results for MD 1, 2, and 3, respectively. The columns show the results for 2D US (left) and 3D US measurements (right).
Table 1.
Mean and standard deviation (mean±SD) in percent (%) of the intraobserver variability for each MD in 2D and 3D for both groups (Gr. 1 and 2).
The last column shows whether a significant difference exists between the averaged 2D and 3D volume estimation sets, being the significant different p-values in bold.
Fig 7.
Bland-Altman plots for the interobserver variability between sets of two medical doctors from Group 2.
All values are given in ml. Row 1, 2 and 3 show results between MD 1 and MD 2, MD 1 and MD 3, and MD 2 and MD 3, respectively. The columns show the results for 2D (left) and 3D measurements (right).
Table 2.
Mean and standard deviation (mean±SD) in milliliters (ml) of the interobserver variability of two MDs in 2D and 3D for both groups.
Significant different p-values are marked in bold.
Table 3.
Mean and standard deviation (mean±SD) of the volume of the first scan series for each MD in 2D and 3D in milliliters (ml).
Statistical significance with respect to the MRI volume is reported, being statistical significant p-values in bold.