1 Introduction
2 Methods
2.1 MOCAIP algorithm
2.1.1 Detection, clustering, and validation
2.1.2 Detection of candidate peaks
2.1.3 Assignment of detected peaks
2.2 Regression analysis for peak designation
2.2.1 ICP pulse pre-processing
2.2.2 Prediction assignment algorithm
2.2.3 Regression analysis
2.2.4 Multiple linear regression
2.2.5 Spectral regression analysis
2.2.6 Support vector machine regression
2.2.7 Extremely randomized decision trees
3 Results
3.1 Comparative analysis of regression methods
3.1.1 Prediction accuracy
3.1.2 Prediction accuracy on new patients
3.1.3 Number of training samples
3.2 Peak recognition
Actual value | Prediction P1 | Prediction P2 | Prediction P3 | |||
---|---|---|---|---|---|---|
Peak | No peak | Peak | No peak | Peak | No peak | |
Peak | 11,223 | 23 | 12,542 | 109 | 12,663 | 129 |
No peak | 86 | 1,482 | 8 | 155 | 6 | 16 |
P
1 (%) |
P
2 (%) |
P
3 (%) | |
---|---|---|---|
KSR (global) | 99.37 | 99.42 | 99.59 |
KSR (individual) | 99.15 | 99.09 | 98.95 |
Gaussian Priors | 97.26 | 92.84 | 90.83 |