Professional Documents
Culture Documents
5000308
ORIGINAL ARTICLE
School of Mechatronics Engineering and Automation of Shanghai University, Shanghai Key Laboratory of Power Station
Automation Technology, Shanghai - PR China
In this paper, a novel artificial anal sphincter (AAS) system based on rebuilding the rectal sensation
function is proposed to treat human fecal incontinence. The executive mechanism of the traditional
AAS system was redesigned and integrated for a simpler structure and better durability. The novel
executive mechanism uses a sandwich structure to simulate the basic function of the natural human
anal sphincter. To rebuild the lost rectal sensation function caused by fecal incontinence, we propose
a novel method for rebuilding the rectal sensation function based on an Optimal Wavelet Packet Basis
(OWPB) using the Davies-Bouldin (DB) index and a support vector machine (SVM). OWPB using a DB
index is used for feature vector extraction, while a SVM is adopted for pattern recognition.
Furthermore, an in vitro experiment with the AAS system based on rectal sensation function rebuild-
ing was carried out. Experimental results indicate that the novel executive mechanism can simulate
the basic function of the natural human anal sphincter, and the proposed method is quite effective for
rebuilding rectal sensation in patients.
Keywords: Artificial anal sphincter system, Executive mechanism, Rectal sensation function rebuilding,
Optimal Wavelet Packet Basis, Support vector machine
Accepted: January 10, 2014
simulate the normal physiology of the human anorectum. highly integrated and the fact that there is no wire linking
The detailed structure model of the executive mechanism the implanted and outside components ensures that the
is discussed, and then its performance characteristics are implantation is less risky and has fewer complications.
assessed with a measurement experiment we established Also, the system comprises three key modules, including
to ensure reliability. the sensor and executive module, the transcutaneous en-
Furthermore, a method for rectal sensation function re- ergy transmission module, and the wireless communica-
building based on optimal wavelet packet (OWP) and sup- tion module as shown in Figure 1.
port vector machine (SVM) is proposed to predict the urge The sensor and executive module comprises a sensor unit
to defecate. Finally, an in vitro experiment of the novel AAS and a novel executive mechanism. The sensor unit placed
system based on rectal sensation function rebuilding is closely to the anal canal acts in the role of the natural nerve
carried out. Experimental result indicates that the proposed system, which has always monitored the pressure signal of
method is quite effective to predict the urge to defecate. the anal canal and transmitted the signal to the inside con-
troller. The novel executive mechanism which simulates
the basic function of human nature anal sphincter is used
SYSTEM OVERVIEW to prevent unwanted leakage of feces.
Driving energy for the implanted component is provided
The novel AAS system proposed in this paper is based by transcutaneous energy transmission module based on
on rectal sensation function rebuilding to treat fecal in- electromagnetic coupling. Two copper coils separated by
continence. The system consists of two components: the the skin are facing each other closely to transmit energy
one, implanted in the body of patients, is called the im- in the maximal degree. Then some regularization must be
planted component, while the other, set outside the body, processed on the initial energy to satisfy the power stan-
are the outside components. The two components are dard for the whole implanted component.
p
where nj, p is the coefficient length of subspace W j . So the tion ability of feature space (17-19). Hence, the DB index is
original wavelet packet coefficients mean feature vector adopted in this paper to measure the classification ability
set is and to select the optimal wavelet packet basis.
The DB index is based on the scatter matrices of the data.
M = {MEAj, 0, MEAj, 1,...,MEAj, p}, P = 2j – 1.[2] The DB index is obtained through determining the worst
case of separation for each cluster and averaging these
2. The relative energy Pj, p in every subspace is adopted as values as follows (20):
the original relative energy feature set N.
1 K
DB = ∑ max Rij ,
K i =1 i ≠ j
[7]
The energy of the pth subspace occurring at the jth de-
composition level is where K is the number of clusters to distinguish and clus-
ter-to-cluster similarity is:
2
E j , p = ∑ (d jp ( k )) . [3]
k ( Dii + D jj )
Rij = , [8]
Dij
The total energy of the jth decomposition level is
where Dii and Dj j denotes the dispersions of cluster i and
TE j = ∑ E j , p . [4] cluster j respectively, Dij is the distance between the mean
p
value of cluster i and cluster j. The value can be obtained
The relative energy Pj, p in every subspace is through the following formula:
1/ 2
1 Ni
∑
2
Pj , p = E j , p / TE j . [5] Dii = y n − mi [9]
Ni n=1
The original relative energy feature set is and Dij = mi − mj , [10]
1 n
F= (w·w ) + C ∑ ξi , [11]
2 i =1 Fig. 7 - Characteristic curve of the electromagnetic and spring forces.
a b
Fig. 8 - Coefficients mean of each subspace. a) Three kinds of Class 1 signal. b) Three kinds of Class 2 signal.
the electromagnetic force increases as the moveable core indicates the urge to defect. The samples without HAPC
approaches the stationary core and the two limit values signal were regarded as Class 2, which means no need to
obviously differ. When the working displacement x is be- prepare for defecation.
tween 4~10 mm, the characteristic curve F − x is approxi- The Db3 wavelet with the best regularity properties was
mately horizontal, which means a constant force. chosen as the mother wavelet to decompose the rectal
The clamping force imposed on the rectum is created by pressure signal. Three-layer wavelet packet decomposi-
the spring compressive deformation, so an increase in in- tion was operated on the rectal pressure signal and 15 dif-
p
testinal contents results in extra compressive deformation, ferent subspaces Wj were obtained. Then the coefficient
which in turn generates an extra force to keep the closing mean MEAj, p and relative energy Pj, p in each subspace
state. When the electromagnet is working at 12 V voltage was obtainable according to Formula [1] and Formula [5]
with maximum current 0.15 A, the electromagnet tempera- respectively, as shown in Figures 8 and 9.
ture rises about 30°C within a continuous 30-minute pe- We adopted three kinds of class 1 signal and three kinds
riod, which causes no harm to human tissues. Repeated of class 2 signal to extract feature vector to predict the
tests indicate that the sealing effect is good enough to urge to defecate. The feature vector of the Class 1 signal
maintain the pressure and prevent feces leakage. is shown in Figure 8a and Figure 9a; the feature vector of
the Class 2 signal is shown in Figures 8b and 9b. As the
Experiment for rebuilding rectal sensation DB index algorithm is based on the entire wavelet pack-
et decomposition tree, 14 MEAj, p and 14 Pj, p should be
The biological parameters telemetry capsule made by calculated. So M is a 14 dimension feature vector of the
Shanghai Jiaotong University was used for rectal pressure subspace coefficient mean and N is a 14 dimension fea-
signal acquisition of 10 healthy volunteers, including six men ture vector of the subspace relative energy. As is shown
and four women. One hundred groups of data with 10 times in Figures 8 and 9, the subspace number 1, 2, 3,..., 14
per person were acquired. Four wrong groups were aban- denotes the subspace W10 ,W11,W20 ,...,W37 respectively. So
doned, while the remaining 96 groups of data were filtered the original feature vector set X = {M, N} can be obtained,
and de-noised. Then 62 groups of data were selected as where X is a 28 dimension vector.
training samples and 34 groups of data as test samples. The The DB index of each subspace in each layer is calculated
samples with HAPC signal were regarded as Class 1, which according to Eq. [7], and then selection of optimal wavelet
a b
Fig. 9 - Relative energy of each subspace. a) Three kinds of Class 1 signal. b) Three kinds of Class 2 signal.
packet basis is extracted with the optimal wavelet packet TABLE I - PREDICTION ACCURACY OF SUPPORT VIRTUAL
basis search algorithm based on the DB index (see above MACHINE
“Optimal Wavelet Packet Basis based on Davies-Bouldin
Predication Kernel Test sample Correct Predication
index”). One optimal wavelet packet basis selection pro-
model function number predication accuracy
cess is shown in Figure 10. The value in each subspace is
the DB index value of the subspace. The colored subspace SVM RBF 34 33 97.06%
is the optimal wavelet packet basis based on DB index, that SVM = support virtual machine; RBF = radial basis function.
is W20 , W32 , W33 , W34 , W35 and W23 . Based on the subspaces
corresponding to the selected optimal wavelet packet ba-
sis, the MEAj, p and Pj, p which are not belong to the sub- SVM. When the HAPC signal is detected in any group of
spaces from feature vector M and N are removed. The goal test data, it indicates occurrence of the urge to defecate.
to simplify the original feature vector set X = {M, N} is thus The prediction accuracy of the SVM with 34 groups of test
achieved. sample is shown in Table I. The test prediction accuracy is
In the end, the simplified feature vector X′ = {M′, N′} of 97.06% and the operating time is 2.13 s.
62 groups of training data is inputted into the SVM for According to the subspaces corresponding to the opti-
model training. Then the 34 groups of test data are input- mal wavelet packet basis based on THE DB index, the
ted to the SVM to testify the effectiveness of the proposed coefficients mean and relative energy in each subspace
algorithm. The RBF kernel is used in this paper for the was selected as a feature vector to predict the urge to
a b
Financial Support: This work was supported by National Natural Address for correspondence:
Science Foundation of China (No. 31100708, No. 31370998) and Hua Wang
Natural Science Foundation of Shanghai (No.11ZR1412400). School of Mechatronics Engineering and
Automation of Shanghai University
Shanghai Key Laboratory of Power Station Automation Technology
Shanghai 200072, PR China
Conflict of Interest: None. wanghua6091@163.com