Skip to main content

Table 4 Comparison of DaTscan image feature extraction methods

From: A three-stage, deep learning, ensemble approach for prognosis in patients with Parkinson’s disease

Method MAPE MAE MSE r R2
Ensemble approach 18.36%
(11.74%, 24.98%)
4.70
(3.56, 5.84)
34.53
(18.81, 50.25)
0.84 0.71
DaTscan + Semi-quantitative + All ImageNet 19.64%
(12.09%, 27.18%)
4.79
(3.65, 5.94)
35.48
(20.11, 50.85)
0.82 0.67
DaTscan + Semi-quantitative 18.82%
(11.32%, 26.31%)
4.74
(3.36, 6.12)
40.60
(14.59, 66.60)
0.79 0.62
DaTscan + All ImageNet 20.64%
(11.50%, 29.79%)
4.81
(3.45, 6.18)
40.95
(19.94, 61.96)
0.79 0.63
Semi-quantitative + All ImageNet 18.82%
(12.75%, 24.89%)
4.83
(3.69, 5.96)
35.57
(20.41, 50.73)
0.82 0.67
DaTscan 19.89%
(12.48%, 27.30%)
5.04
(3.78, 6.29)
40.41
(22.09, 58.74)
0.81 0.66
Semi-quantitative 21.43%
(13.76%, 29.09%)
5.58
(4.13, 7.03)
51.29
(27.25, 75.33)
0.76 0.57
All ImageNet 20.18%
(13.76%, 26.59%)
5.15
(3.90, 6.40)
41.44
(20.15, 62.72)
0.78 0.62
VGG16 21.67%
(13.99%, 29.35%)
5.54
(4.14, 6.94)
49.31
(26.95, 71.67)
0.75 0.56
ResNet50 19.70%
(13.82%, 25.58%)
5.32
(4.03, 6.61)
44.14
(25.64, 62.65)
0.79 0.62
DenseNet121 20.24%
(13.04%, 27.45%)
5.23
(3.90, 6.56)
44.12
(23.23, 65.01)
0.79 0.63
InceptionV3 21.89%
(14.30%, 29.48%)
5.47
(4.17, 6.76)
45.89
(25.32, 66.46)
0.76 0.58
  1. Data in parentheses are 95% confidence intervals
  2. All values for r were significant (P < 0.001)
  3. MAPE, mean absolute percentage error; MAE, mean absolute error; MSE, mean squared error; r, Pearson correlation coefficient; R2, coefficient of determination