Table 2.
Overview of results
| A | ||||||
|---|---|---|---|---|---|---|
| Data set | Noise | Scale | Otsu’s | Cutoff 6 | Cutoff 8 | Cutoff 10 |
| D2 | BI | 1 | 99.4% | 95.0% | 98.0% | 98.7% |
| 2 | 97.6% | 93.7% | 95.5% | 96.7% | ||
| 3 | 93.8% | 92.0% | 93.4% | 94.6% | ||
| 4 | 90.3% | 90.3% | 91.2% | 92.7% | ||
| 6 | 85.8% | 90.0% | 90.8% | 91.5% | ||
| Sc | 10 | 98.5% | 88.4% | 93.6% | 97.9% | |
| 20 | 88.7% | 87.9% | 93.0% | 96.9% | ||
| 30 | 81.2% | 87.7% | 92.2% | 93.5% | ||
| 40 | 76.7% | 87.4% | 90.5% | 87.6% | ||
| 50 | 74.3% | 86.5% | 88.0% | 82.3% | ||
| 60 | 73.2% | 85.6% | 86.3% | 80.5% |
| B | |||||
|---|---|---|---|---|---|
| Data set | Validation | M | Cutoff 6 | Cutoff 8 | Cutoff 10 |
| D3 | DS | Gold standard |
97.0% ± 0.7 range 95.6–97.9 |
96.6% ± 0.8 range 95.0–97.7 |
95.9% ± 0.9 range 93.8–97.2 |
| VE |
16.7 mm3 ± 5.5 range 8.8–30.7 |
11.5 mm3 ± 5.7 range 5.0–25.5 |
17.1 mm3 ± 7.4 range 8.4–33.6 |
23.3 mm3 ± 8.7 range: 13.0–41.0 |
|
| VE/M | 1 |
0.7 ± 0.2 range 0.4–0.9 |
1 ± 0.2 range 0.7–1.5 |
1.4 ± 0.3 range 1.0–2.1 |
|
| VT | 276.2 mm3 ± 37.6 (range 223.6–347.6) | ||||
As an artificial data set, D2 provided a known ground truth to test and compare VOLT cutoff versions to Otsu’s method (O). A shows an overview of the Dice scores (DS) of each segmentation method (Otsu’s, cutoff 6, cutoff 8, cutoff 10) concerning the real-world MRI imitating noise that was added stepwise in the form of increasing blurriness noise (Bl, Gaussian blur kernel, SD range 1–6 voxel in x/y/z-direction; SD = standard deviations) or increasing scatter noise (Sc, SD range of intensity variation: 0–50 SD). For visualization of the added noise and results, see Fig. 1a. D3 included real-world data sets from consecutive patients from the interdisciplinary German Center for Vertigo and Balance Disorders, Munich, Germany. Part B shows an overview of the results’ mean of each segmentation method (manual segmentation that was considered as the gold standard and VOLT with three different cutoffs 6, 8, 10). Segmentation accuracy was evaluated using the Sørensen-Dice overlap coefficient, and segmentation precision were estimated by comparing the volume of the ELS (VE). The ratio VE/M was supplied to show the deviation of each cutoff from the gold standard, which was the manual segmentation. The VE ranges include all different grades of endolymphatic hydrops
± standard deviation, Bl blurriness, DS Dice score, Sc scatter, VE volume of the endolymphatic space, VT volume of the total fluid space