Fig 1.
N: normal people; P: CHF patients, in which 1 is of NYHA I-II, 2 is of NYHA III, 3 is of NYHA III-IV; S1: basic measures of 24-h RR interval data, which reflect long-term data variation); S2: basic measures of the second 5-min segment, which representing a stable measurement condition of short-term data; S3: mid-value of basic measures of 5-min segments, which showing an intermediate state of short-term data; D1: mean value of basic measures of 5-min segments, for robustness improvement; D2: standard deviation of each basic measure of 5-min segments; D3: root mean square of each basic measure of 5-min segments; D4: coefficient variation of each basic measure of 5-min segments; D5: percentage of abnormal value (value intervening M±S) of each basic measure of 5-min segments; D6: sample entropy of each basic measure of 5-min segments; D7: fuzzy entropy of each basic measure of 5-min segments.; DT-SVM: decision tree based support vector machine.
Fig 2.
Multistage classification algorithm based on DT-SVM for risk assessment.
Upper diagram: tree-structured classifier. Lower diagram: wrappers for feature selection. N: normal samples; P: CHF patients, in which 1 is of NYHA I-II, 2 is of NYHA III, 3 is of NYHA III-IV; DSF: disease screening function; RAF: risk assessment function, in which I is for discriminating the higher risk from the lower risk, II is for distinction of moderate risk and mild risk; BE: backward elimination; SD: significance difference.
Table 1.
Classification performance of classical SVM in 4-level risk assessment.
Table 2.
Performance of different feature combinations for disease detection and quantification.
Table 3.
Result of node selection for level 1 among all samples.
Table 4.
Result of node selection for level 2 among CHF patients.
Fig 3.
Multistage risk assessment model of CHF.
DSF: disease screening function to detect normal from patients; RAF: risk assessment function, in which I is for discriminating the higher risk from the lower risk, II is for distinction of moderate risk and mild risk; N: normal samples; P: CHF patients, in which 1 is of NYHA I-II, 2 is of NYHA III, 3 is of NYHA III-IV.
Table 5.
Selected optimal feature subsets for each level with backward elimination.
Fig 4.
N: normal samples; P: CHF patients, in which 1 is of NYHA I-II, 2 is of NYHA III, 3 is of NYHA III-IV.
Table 6.
Classification performance.
Table 7.
Highlight.