Hybrid Correlation-Neural Network Synergy for Gait Signal Classification - Advances in Heuristic Signal Processing and Applications - page 270

Fig. 12.8 Flowchart representation of scheme [ 13 ]

lem where we determine whether the subject tested is healthy/pathological. If the
subject is diagnosed as pathological, then ERNN2 is activated with the same set of
feature vectors, determined for that specific subject. ERNN2 is specifically trained
with the training data set determined from pathological subjects only, and it is de-
signed to solve a three-class problem, segregating pathological subjects into PD,
HD, and ALS classes. Figure 12.8 shows Scheme 1 in a flowchart form. Finally,
outputs from both ERNN1 and ERNN2 are utilized to suggest the ultimate diagno-
sis which classifies the subject under consideration into one among the four classes,
i.e., healthy/PD/HD/ALS.
The same problem can also be solved by employing Scheme 2, shown in
flowchart form in Fig. 12.9 . The feature extraction part remains identical with that
of Scheme 1, but the classification module now employs three modular ERNNs,
namely ERNN1, ERNN3, and ERNN4, each trained to perform specific binary
classification jobs. ERNN1 is implemented in an identical manner with that of
Scheme 1. But if it diagnoses the subject as pathological, then ERNN3 is activated
to determine whether the pathological subject is suffering from ALS or not. If the
answer is negative, then ERNN4 is activated to determine whether the subject is
suffering from PD or HD. The final outcome of the automated tool discussed in
Scheme 2 is determined by considering outputs from all three ERNNs.