njit-etd2003-081 - New Jersey Institute of Technology
njit-etd2003-081 - New Jersey Institute of Technology njit-etd2003-081 - New Jersey Institute of Technology
233 Table 5.9 Principal Components of Different Population Groups PC1 PC2 PC3 Normal (Exercise) Normal + COPD1 RSP BP LF_PCOH_HR_BP RSP HF_COH_HR_BP HF_PCOH_HR_BP Normal + COPD2 LF _ COH _ HR _ RSP HF' _ COH _ BP _ RSP RSP Table 5.9 shows the summary of the principal component analysis for the three cases of the HRV study for both normal and COPD subjects. The data set for normal subjects during exercise shows the respiration rate was first principal component (PC); the blood pressure was the second PC and the partial coherence value between heart rate and blood pressure in the low frequency range was the third PC. When the first set of data that combined the COPD and normal data was used, only the first PC of respiration was the same and the other two PC's were different. When the second set of the combined COPD-normal data was used, the three PC's were completely different. This fact indicates that the PCA method is not "event" dependent but the technique is only adaptive to the actual data used.
234 5.7 Cluster Analysis The purpose of using principal component analysis (PCA) and cluster analysis (CA) is illustrated by the results of figures 5.63 and 5.64 and summarized in Table 5.10. First, PCA-CA was used to blindly separate the normal subjects and COPD patients using a data set that contained the physiological data such as heart rate, blood pressure rate and respiration rate as well as the cross-spectral results of the (weighted) coherence and partial coherence in both the LF and HF range. Figure 5.63 clearly shows the separation between the COPD group (green squares and blue pluses) from the normal group (red circles). In the same figure, one red circle of a normal subject was placed near a blue plus and a green square of the COPD subject. The reason for this may be that the COPD subjects represented by blue plus and green square may only be at risk of the disease. Hence, their physiological parameters placed them near that of the normal subject. The COPD group is represented in green and blue markers because their data came from the same COPD group, but were from two different testing trials. This proves that cluster analysis could separate the normal and COPD subjects and the results were reproducible, i.e. same results for the same COPD group even with different data sets from different experimental trials. Figure 5.64 shows the results of not only normal and COPD blind separation but also the COPD severity classification. When the desired number of clusters was set prior to running the cluster analysis program to separate the normal group (cluster 1) from the "at risk" group (cluster 2), the "mild' group (cluster 3), the "moderate" group (cluster 4) and the "severe' group (cluster 5), the result was the graph of figure 5.64. This is the first attempt at this type of blind separation and classification. Another more fine-tuned
- Page 211 and 212: 182 These figures basically show th
- Page 213 and 214: 184 Figure 5.20 Sympathetic and par
- Page 215 and 216: 186 Figure 5.24 Sympathetic and par
- Page 217 and 218: 188 5.2.7 Time-Frequency Analysis (
- Page 219 and 220: 190 Figure 5.29 3D and contour plot
- Page 221 and 222: 192 Figure 5.33 3D and contour plot
- Page 223 and 224: 194 Figure 5.34 Sympathetic and par
- Page 225 and 226: 196 Figure 5.38 Sympathetic and par
- Page 227 and 228: 198 Figure 5.42 Sympathetic and par
- Page 229 and 230: Figure 5.44 Plot of raw respiration
- Page 231 and 232: Figure 5.46 The LF partial coherenc
- Page 233 and 234: Figure 5.48 HF partial coherence pl
- Page 235 and 236: Table 5.2 Cross-Spectral Analysis o
- Page 237 and 238: Table 5.3 Cross-Spectral Analysis o
- Page 239 and 240: Figure 5.50 HF coherence of COPD (1
- Page 241 and 242: 212 For better presentation of the
- Page 243 and 244: Figure 5.53 Coherence and partial c
- Page 245 and 246: 216 2. Interpretation of the transf
- Page 247 and 248: 218 covariances of the parameters,
- Page 249 and 250: 220 deviations are interpreted as A
- Page 251 and 252: 222 Figure 5.58 Bode plot of the HR
- Page 253 and 254: 224 In this section a simple model
- Page 255 and 256: 226 The data for all 47 COPD subjec
- Page 257 and 258: 228 Figure 5.60 Normal and COPD cla
- Page 259 and 260: 230 Figure 5.61 Normal and COPD cla
- Page 261: 232 Figure 5.62 Normal classificati
- Page 265 and 266: 236 Figure 5.64 Severity classifica
- Page 267 and 268: 238 both the normal and COPD subjec
- Page 269 and 270: 240 In summary, COPD subjects had h
- Page 271 and 272: APPENDIX A EXERCISE PHYSIOLOGY A.1
- Page 273 and 274: 244 A.3 Figure Out Your Target Hear
- Page 275 and 276: APPENDIX B ANALYSIS PROGRAM LISTING
- Page 277 and 278: 248 4) Click on file, close to exit
- Page 279 and 280: 250 • TN 11
- Page 281 and 282: 252 B.1.2 Partial Coherence Between
- Page 283 and 284: 254
- Page 285 and 286: 256 Block Diagram !rime of record K
- Page 287 and 288: 258
- Page 289 and 290: 260 B.2.2 Time — Frequency Analys
- Page 291 and 292: 262 This program provides the STFT
- Page 293 and 294: 264 G(:j+1)=G(:,j+1)/(2*sum(G(:j+1)
- Page 295 and 296: 266 T=(length(Signa)/sample)/(Times
- Page 297 and 298: 268 subplot(3, 1,3), plot(T,E); xla
- Page 299 and 300: 270 4. The program creates five out
- Page 301 and 302: 272 B.2.3.4 Program to Generate Sym
- Page 303 and 304: 274 ylabel('frequency'); title('Ins
- Page 305 and 306: 276 The program will run and output
- Page 307 and 308: 278 axis([0 1 0 2]); grid on; xlabe
- Page 309 and 310: 280 vagal=sum(TFDs(HFC,1:k)); symto
- Page 311 and 312: 282 plot(J,symtopar); %plot(A,symto
234<br />
5.7 Cluster Analysis<br />
The purpose <strong>of</strong> using principal component analysis (PCA) and cluster analysis (CA) is<br />
illustrated by the results <strong>of</strong> figures 5.63 and 5.64 and summarized in Table 5.10. First,<br />
PCA-CA was used to blindly separate the normal subjects and COPD patients using a<br />
data set that contained the physiological data such as heart rate, blood pressure rate and<br />
respiration rate as well as the cross-spectral results <strong>of</strong> the (weighted) coherence and<br />
partial coherence in both the LF and HF range. Figure 5.63 clearly shows the separation<br />
between the COPD group (green squares and blue pluses) from the normal group (red<br />
circles). In the same figure, one red circle <strong>of</strong> a normal subject was placed near a blue plus<br />
and a green square <strong>of</strong> the COPD subject. The reason for this may be that the COPD<br />
subjects represented by blue plus and green square may only be at risk <strong>of</strong> the disease.<br />
Hence, their physiological parameters placed them near that <strong>of</strong> the normal subject. The<br />
COPD group is represented in green and blue markers because their data came from the<br />
same COPD group, but were from two different testing trials. This proves that cluster<br />
analysis could separate the normal and COPD subjects and the results were reproducible,<br />
i.e. same results for the same COPD group even with different data sets from different<br />
experimental trials.<br />
Figure 5.64 shows the results <strong>of</strong> not only normal and COPD blind separation but<br />
also the COPD severity classification. When the desired number <strong>of</strong> clusters was set prior<br />
to running the cluster analysis program to separate the normal group (cluster 1) from the<br />
"at risk" group (cluster 2), the "mild' group (cluster 3), the "moderate" group (cluster 4)<br />
and the "severe' group (cluster 5), the result was the graph <strong>of</strong> figure 5.64. This is the first<br />
attempt at this type <strong>of</strong> blind separation and classification. Another more fine-tuned