Fig 1.
Overall method flow chart.
Fig 2.
EMD decomposition principle.
Fig 3.
The EMD decomposition of heart sound.
(a) Time domain; (b) Frequency domain.
Fig 4.
The flowchart based on EMD with adaptive reconstruction.
Fig 5.
Heart sound flowchart based on EMD adaptive HD reconstruction.
Fig 6.
Correlation and RMSE of IMFs of heart sound.
(a) Correlation and RMSE of IMF of AS, MS, MR, MVP; (b) Correlation and RMSE of IMFs of ASD, VSD, TOF&NHS.
Fig 7.
Adaptive selection based on the mean of the first 7 layers of HD.
(a) Adptive Hausdorff Distance of IMFs of AS, MS, MR, MVP; (b) Adptive Hausdorff Distance of IMFs of ASD, VSD, TOF&NHS.
Fig 8.
Time-frequency of heart sound EMD reconstruction.
Table 1.
40 fusion features.
Table 2.
Types of samples in open source [25].
Table 3.
Types of samples in our laboratory.
Fig 9.
The average classification accuracy versus the number of top-ranking features selected by mRMR, KCCAmRMR, QPFS, MIC and Tree and RFECV using RF classifier from open source data.
Fig 10.
The average classification accuracy versus the number of top-ranking features selected by mRMR, KCCAmRMR, QPFS, MIC and Tree and RFECV using RF classifier from our Lab data.
Table 4.
Comparison of accuracy and feature dimension under different methods based on mRMR.
Table 5.
Comparison of accuracy and feature dimension under different methods based on KCCAmRMR.
Table 6.
Comparison of accuracy and feature dimension under different methods based on QPFS.
Table 7.
Comparison of accuracy and feature dimension under different methods based on MIC.
Table 8.
Comparison of accuracy and feature dimension under different methods based on tree.
Table 9.
Comparison of accuracy and feature dimension under different methods based on RFECV.
Table 10.
The best accuracy and number of selected minimum features by adaptive HD method.
Table 11.
Comparison of results using Yaseen database.