Table 1.
Quantitative analysis of segmentation results obtained from volume shown in Figs. 1 and 2. Segmentation results obtained from DeepSynth were quantitatively compared with those obtained from FARSIGHT, Squassh and CellProfiler using either default settings or settings optimized as described in “Methods”.
| Segmentation technique | Time (entire volume) | Voxel based | Object based | ||||
|---|---|---|---|---|---|---|---|
| Type I | Type II | Accuracy | Precision | Recall | F1 | ||
| Sub-volume collected 75–112 microns from the surface | |||||||
| DeepSynth | 94 sec | 4.03% | 3.81% | 92.15% | 72.80% | 90.55% | 80.71% |
| FARSIGHT Default | 13 min | 9.61% | 0.92% | 89.47% | 65.94% | 94.62% | 77.72% |
| FARSIGHT Optimized | 13 min | 9.55% | 1.01% | 89.44% | 78.09% | 87.11% | 82.53% |
| Squassh Default | Hours | 9.56% | 0.39% | 90.05% | 92.94% | 33.19% | 48.92% |
| Squassh Optimized | Hours | 11.45% | 0.36% | 88.19% | 90.41% | 27.62% | 42.31% |
| CellProfiler Default | 15 min | 7.15% | 2.02% | 90.83% | 80.12% | 58.37% | 67.54% |
| CellProfiler Optimized | 15 min | 5.36% | 3.06% | 91.58% | 71.04% | 78.89% | 74.76% |
| Otsu-3DWatershed | 54 sec | 8.99% | 1.43% | 89.58% | 90.58% | 52.52% | 66.49% |
| Sub-volume collected 130–162 microns from the surface | |||||||
| DeepSynth | 94 sec | 3.24% | 4.34% | 92.42% | 72.94% | 92.54% | 81.58% |
| FARSIGHT Default | 13 min | 4.07% | 5.05% | 90.88% | 43.18% | 67.86% | 52.78% |
| FARSIGHT Optimized | 13 min | 4.08% | 5.04% | 90.88% | 78.95% | 68.18% | 73.17% |
| Squassh Default | Hours | 8.64% | 2.63% | 88.73% | 83.33% | 35.21% | 49.50% |
| Squassh optimized | Hours | 3.80% | 4.71% | 91.49% | 76.47% | 39.39% | 52.00% |
| CellProfiler Default | 15 min | 1.30% | 7.35% | 91.35% | 55.32% | 48.15% | 51.49% |
| CellProfiler Optimized | 15 min | 0.46% | 10.95% | 88.59% | 28.57% | 26.09% | 27.27% |
| Otsu-3DWatershed | 54 sec | 3.76% | 5.53% | 90.71% | 62.50% | 40.98% | 49.50% |
The values for “Time” reflect the times required to obtain segmentations using. Accuracy was measured using both voxel-based metrics (voxel-by-voxel agreement with ground-truth data) and object-based metrics (agreement in the detection of objects with ground-truth data) in 64 by 64 by 64 voxel sub-volumes obtained 75–112 microns from the surface of the sample (top) and 130–162 microns from the surface of the sample (bottom). For voxel-based accuracy, type-I error (false positive rate) represents the fraction of voxels in the volume wrongly detected as belonging to nuclei and type-II error (false negative rate) represents the fraction of voxels wrongly detected as background. Object-based accuracy is measured using the F1 score, which is the harmonic mean of precision and recall, where precision is the ratio of the number of correctly identified nuclei to the sum of the number of correctly identified nuclei plus the number of objects incorrectly identified as nuclei and recall is the ratio of the number of correctly identified nuclei to the sum of the number of correctly identified nuclei plus the number of nuclei that failed to be detected. Details of the analyses are described in “Methods”.