Scribd Logo
Upload

Welcome to Scribd!

  • 2

    Uploaded by

    Oana Drăgan
    5 views11 pages

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    5 views11 pages

    2

    Uploaded by

    Oana Drăgan

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 11
    Computers in Boley and Medicine 60 (2015) 132 Contents lists available at ScienceDirect Computers in Biology and Medicine & Journal homepage: www elsevier.comilocate/ebm Automatic detection of atrial fibrillation using stationary wavelet transform and support vector machine ‘Shadnaz Asgari**, Alireza Mehrnia”, Maryam Moussavi“ gre computer nse nd emp, calor Sut Une ag ec. 1250 Bone Bar HS 2, Lrg ra " Depertment of Hectic! Engineering Univesity of Cfo Ls Angeles S61258 Engineering IV Bilin, Bo 91554, Ls Angels, CA SOO, CA «Deparment of Eecrncal Engen Calfome tte Unversity Long Beach, 1250 Bower Boulevard Lang Beach CA S840, Ch ARTICLE INFO ABSTRACT Received 19 August 20:4 ‘Accepiod 6 Match 2015 Background: Atrial Abrilation (AF isthe most common cardia arhythmia, and a major public health burden associated with signifcant morbidity and mortal. Automatic detection of AF could substan- tally help in early diagnosis, management and consequently prevention ofthe complications associated With chronic AF. In this paper, we propose a novel method for automatic AF detection ayer fethod: Stationary wavelet transform and suppest vector machine have been employed to detect AF ‘tal bilo episodes. The proposed method eliminates the nced for P-peak or R-Peak detection (a pre-processing Suppor vector machine step required by many existing algorithms) and hence it performance (sensitivity. specificity) does not (ari arta aC curve ana, depend on the performance of beat detection. The proposed method has been compared with those of the existing methods in terms of various measures including performance. transivon time (detection {ela associated with transitioning from 2 non-AF fo AF episode} and computation time (using MIT-BIA ural Fibilation database) iss Results ofa stated 2-foldcros-validtion reveals thatthe area under the Receiver Operative Characteristics (ROC) curve ofthe proposed method is ©9.5%. Moreover, the methed maintains Is high accuracy regardless of the choice ofthe parameters values and even for data segments a shor as 10, Using the optimal values of the parameters, the method achieves sensitivity and speciety of 7.0% and 971%, respectively Discussion: The propesed AF detection method has high sensitivity and specificity, and holds seve interesting properties which make ita stable choice for practical applications. ‘2015 Elsevier Ld All rghts reserved 1. Introduction Atrial fibilation (AF) is the most common cardiac archythmia with an estimated prevalence of 1% corresponding to 2:3 million patients inthe US and 45 million in the European Union [1]. This prevalence is strongly associated with age, such that over 17% of People aged 85 or above are AF patients [2|. tis expected that the number of AF patients in US. will increase by 2.5-fold to more than 56 million people by 2050 |3|. This increase not only reflects the growing population of elderly individuals, but also. improved Survival of people with predisposing conditions to AF (eg. ischemic heart disease) [4]. The presence of AF is associated with a 5-fold increased risk of stroke [5] and about 2-fld increased risk of death, independent of other risk factors (6). Furthermore, due to high rates ‘mal adress: Shdnaz Asgaricslhed (© ASB) ‘npc er 101016 compbiared2015.03.003 (10-25 2015 Beier Ld Al ight esered of hospitalization and considerable health resource utilization, the ‘economic and clinical burden of AF is substantial and will continue ta increase in the future (7) AF is usually characterized by rapidity and irregularity of ventricular contraction, The chronic forms of AF can be parox- yysmal AF (more than one episode with spontaneous termination within seven days); persistent AF (not sef-terminated, or lasted ‘more than seven days) and permanent AF (not terminated, or terminated but relapsed) [9 Silent or asymptomatic AF can also ‘occur in any ofthese temporal forms, carrying a similar prognosis to symptomatic AF [10]. Chronic AF adversely affects the blood flow ‘dynamics and can result in the stoke, Early and accurate detection of AF and its management (eg anticoagulation, antiarrhythmic therapy and radiofrequency ablation) could substantially help in prevention of the complications associated with the chronic AE Currently, the diagnosis of AF mainly rests on the presence of related symptoms (eg. shortness of breath and fatigue), followed by an electrocardiogram (ECG) study to verify the diagnosis. S. Agar t lf Computers in ily and Meine 6 (205) 132-142 3 During the ECG study, a trained clinician/technician visually inspects the ECG signal collected aver short period of time (ea, less than 48 h) to identify the AF associated chaotic patterns ‘or abnormal changes in the waveform morphology. Early and accurate diagnosis of AF isa challenging task. AS a result, a majority of the early stage and easily treatable AF cases are not diagnosed in time and evolves into chronic debilitating AF ‘with high cardiac related complications Several factors hinder the accurate and early detection of AF: ‘© Many of the AF patients do not experience symptoms. Hence, a considerable number of AF cases are undetected or diagnosed fortuitously when the subject is being evaluated for other ‘cardiac complications. Moreover, the studies have shown that 2 good correlation does not always exist between symptoms and episodes of AF (11) ‘© ECG study can miss detection of AF in patients with paroxysmal intermittent AF who are not experiencing AF episodes during the study. ‘Visual inspection of hours of collected ECG is time consuming. Accurate interpretation of ECG study requires extensive train- ing and experience. Hence, the reliability of diagnosis is dependent on the level of training or experience of the clinician. in fact, studies have shown that many primary care practitioners are not able to detect AF (on an ECG) with sufficient accuracy to guide therapy [12] AA possible solution to this problem is automatic AF detection using ambulatory monitoring. This approach could eliminate the heed for visual inspection, and contribute to early and accurate detection of AF by allowing for an in-depth analysis of ECG and identification of abnormal patterns. Over the last two decades, several algorithms have been devel- ‘oped for automatic AF detection, Many of these algorithms either rely on the absence of P-waves (replaced by rapid oscillations or Absillatory waves) [13-16] or R-R irregularities [17-21], ar combi- nation of both characteristics to detect AF episodes [22-24]. Ae Pewaves are prone to contamination with motion and noise artifacts, the AF detection algorithms solely based on the absence ‘of P-waves perform poorly in the presence of noise [25]. Therefore, employing heart rate variability for AF detection has become a preferred approach in recent years. But many ofthese R-R interval based methods compare the density histogram of R-R intervals for 2 segment of data with previously compiled histograms of R-R imervals during AF using the Kolmogorov Smirnov test [18.24.26] Since these algorithms requie storage of large amount of histogram data, they may not be suitable for implementation in an ambulatory ‘monitoring device with limited memory and processing, pawer (Other R-R algorithms need various parameters tuning to guarantee the high accuracy of AF detection [19-21] Furthermore, most ofthe existing methods do not perform well with short data segments (less than one minute) [25]. This deficiency may result in missing short-duration AF episodes (prevalent in Paroxysmal or early stage AAP less accurate calculation of AF burden (a measure ofthe percent of time a patient spends in AF), and a lower speed of AF detection, Lasty, a majority of existing approaches for AF detection require P (oF R peak detection asa pre-processing step and consequently their performance will degrade if the related peaks are missed or erroneously detected, ‘Aimed at addressing the existing issues, the current work proposes a novel method for automatic AF detection using sta- tionary wavelet transform (SWT) and support vector machine (SVM). As opposed to traditional methods, the proposed method does not require P or R peak detection pre-processing step. Only few parameters are involved in the development of the proposed ‘method and the method maintains its high accuracy regardless of the choice of the parameters! values. The proposed method also performs well even for short data segments of 10s. In what follows, ve frst describe (Section 2) our proposed method in deta Then in Section 3, we present the result of applying the proposed method fon MIT-BIH Atal fibrillation database, These results are discussed and compared with those of the existing methods in Section 4, Conclusions are presented in Section 5 2. Materials and methods 21. Automatic AF detection using wavelet transform and support vector machine ‘The proposed method consists of three major steps: pre- processing: feature extraction; and AP classification, Fig. 1 ilus- tuates a block diagram of various steps of this method. In the following sub-sections, we will describe each step in detail 21. Pre-processing ‘The incoming ECG signal is divided every T seconds into segments of length T. An elliptical band-pass filter with pass- band of 05-50 Hz (effective filter order of 10) is applied to each data segment to remove noise and baseline wander. After filtering in the forward direction, the itered sequence was then reversed and run back through the filter to obtain a zero-phase distortion, ‘Then a wavelet transform is performed on the filtered data segments ‘Wavelet transform has proved to be a useful tool for denoising, Aelineation and compression of signals. Using wavelet transform, fone can observe a signal at different scales where each scale emphasizes on various signal properties and characteristics (27) Wavelet transform allows for the analysis of transients, aperiodi- city and other non-stationary signal features where, through the Femur Bxracton Fig 1 Modular tamenatt of he propsed method fr AF detection. 14 Asari e ob Computers in Blog and Medan 60 (2015) 152-142 omy He ig. 2. Teestructure ofan Lieve story wavelet tansorm (= 1.2, interrogation ofthe transform, subtle changes in signal morphol ‘oay may be highlighted over the scales of interest [28 In the current work, we employ wavelet transform which allows for arbtrariiy high resolution of the high frequency siznal ‘components (in contrast to short term Fourier transform) and then extract features that would enable us to detect AF episodes. ‘Among several wavelet transform techniques, we choose station- ary wavelet transform (SWT) SWTT is time-invariant and at each, decomposition level, its wavelet coefficients carry the same number of samples (temporal information) as that of the original signal. SWT overcomes the problem of repeatability and robust- ress of the analysis which exists with a discrete wavelet trans- form. We leverage these remarkable properties of SWT, and employ it forthe purpose of AF detection. is known that to implement SWT with L-levels on a signal, the length of signal should be a multiple of In order to meet this data length requirement, we zero-pad the filtered data segment (Let us call the zero-padded signal. xin) where 1...) Fig. 2 presents a tree-struture diagram of an I-level SWT (21,2, 1m this figure, G2) and H(z) are high-pass and low-pass filters respectively, designed based on the wavelet basis function, and Xe) isthe zero-padded signal in z-domain. Note that at decom- Position level é, the impulse responses of high-pass filter Giz") {nd low-pass iter") ae up-sampled versions (by a factor oF 2) of impulse responses of high-pass filter Gi2*™*) and low-pass filter #12”) at previous level (¢—1. The detail coeicients dn) and coarse coeicients cin) in time domain can be recursively obtained as Dee t2*")De—1(2) => det) = Lginpde—(n—2°"m), Caz) HC 2) etn) = SM (0-2, N a, Hence, at the end of pre-processing step, a total of 2 x L time series are generated (L detail coefficients and L coarse coefficients), ‘where each coefficient time series has the same time resolution as ‘the original signal x(n), Note that the optimal value forthe length of the data segment (7) is not a trivial choice. The parameter tuning to maximize the performance ofthe proposed method will be discussed later in Section 23. 212. Feature extraction Following the calculation of wavelet coefficients at ths step, 1wo features are extracted for each coefficient: Peak-to-average power ratio: and log-energy entropy. These features are later used in combination with a support vector machine classifier to detect AF. 2.1.2.1, Peaketo-average power ratio. The power spectrum of the ‘wavelet coefficients at each level provides spectral information of ECG signal for different scales (25), and hence it can be used to analyze atrial activity. In fact, power spectrogram of detail Coefficient at th level (9%(f)) and that of coarse coefficient (SEU) can be expressed as, Si) = Et1DAN!7), Sh=BuiCan. where E denotes expectation operation, and D,¢f) and Cf) are Fourier transforms of detail coefficient (Dr(f) =D/2)\< us) and coarse coefficient (Cf) = C2)» = ex, respectively. It is known that atrial activity usually occurs in the frequency range of F=[4-9H2]. Therefore during an AF episode, one can expect to have higher power concentration in this frequency range [25], Peak-to-average power ratio has been shown to be effective in capturing the power distribution profile in the aforementioned frequency range [25], and is obtained according to the following formula 2 ear @ P= Ta where p? and pf are Peak-to-average power ratio of detail and coarse coefficients at level, respectively. By calculating p? and p= for all levels ¢=(1,2,..L}, we obtain 2x1 power spectrum related features for each data segment. 2122. Log-energy entropy. Entropy is a measure of the degree of disorder, uncertainty or randomness of a signal. Wavelet entropy provides useful information about the underlying dynamic system (associated with a signal) over different frequency bands. For example, entropy of a mono-frequency signal is very low. given that the wavelet coefficient atthe decomposition level containing the representative frequency will contain mast of the signal energy. In contrast, a random white noise has high entropy, as there is a significant contribution of energy from wavelet Coefficients at various decomposition levels (30). During AF, electrical discharges conducted from the atrium into the ventricles are irregular and disorganized. Consequently, one can expect that the ECG signal demonstrates more degree of randomness and a higher entropy during an AF episode [30]. Leveraging this note, in order to reveal underlying dynamic process relevant to AF pathology, we calculate the wavelet entropy Of the signal. Various wavelet entropy measures have been defined. In this paper, we use a simple non-normalized log-energy entropy [31] {which is conveniently included in Matlab Wavelet Packet and quickly executable) defined as Sloe dam, a « B= Sloe came, ‘where BP and Ef are the log-energy entropy of detail coefficient and coarse coefficient at éth level, respectively. Calculation of log- tenergy entropy forall wavelet coefficients results in obtaining 2 time related features for each data segment. Hence, atthe end of this step, a vector of 4 features are extracted fer each data segment as PER OP S. Agar t lf Computers in ily and Meine 6 (205) 132-142 ns 213, AF classification ‘Our goal is to classify each data segment into AF or non-AF category using 4L extracted features. For this purpose, we ‘employ support vector machine (SVM) as our classification algo rithm, SVM is a non-parametric binary classifier which has shown promising results in various medical diagnostics [32-35]. Ina conventional problem of binary classification, a data point is viewed as a p-dimensional vector which belongs to one the (wo possible categories. An SVM classifies these data points by finding the best (p- 1)-dimensional hyperplane that separates all data points of ‘one class from those of the other clas, The separating hyperplane has the largest margin between the two classes (Margin is the ‘maximal width of the slab parallel to the hyperplane that has no imerior data points). SVM then classifies new samples based on ‘hich side ofthe hyperplane they fall into and how far the samples are from the Fyperplane (SVM score)(35|, One of the major advan- tages of SVM is that in addition to Iinear classification, it can efficiently perform a non-linear clasifation |37), Using a kemel trick, samples are mapped into higher-dimensional feature space ‘where a linear classifier can separate the two classes with the largest ‘margin among the samples. Mapping the separating hyperplane back to the original space results in having a non-linear classifies For our AF detection, we use a common kernel choice, Gaussian Radial Basis Function kernel Ko, d) 20S, 6 with scaling factor o=1. @ and @; are features of two data Segments and o—d] is the squares Euclidean distance between the two feature vectors. The separating hyperplane for ‘our SVM classifier is obtained using the conventional “Least- Squares” method. 22. Patient data “To evaluate the performance of the proposed method, we ‘employ MIF-BIH Atrial fibrillation (MIF-BIH AFIB) dataset (38, the most popular and most frequently used publicly available ‘dataset fot AF detection. This dataset contains 23 annotated records of ECG signal from atrial fbrillation patients (mostly paroxysmal) with sampling frequency of 250 Hz and 12-bit resolu- tion over a range of ++ 10 mv. Each record is about 10 h-long and the whole dataset includes slightly less than 234 h of data. The ‘dataset has 605 annotated episodes: 291 atrial fibrillation episodes with average time duration of 115 s, 14 arial futer episodes with average time duration of 419 s, 12 episodes of junctional rhythm with average time duration of 27's and 288 episodes of all other sythms with average time duration of 174. Fig. 3 presents examples of ECG signal during four annotated episodes. 23. Data analysis andl validation protocol Daubechies 5 is an orthogonal wavelet resembling the ECG waveform in morphology [25]. Hence, we choose the Daubechies 5 wavelet as the mother wavelet to perform wavelet analysis, Given the sampling frequency of 250 Hz and the frequency range of atrial activities (4-9 Hz) [39], a L=6 level wavelet transform is implemented. A periodogram spectral estimator with a Hamming ‘window is used to obtain the power spectrum of each wavelet coefficient, ‘The origina beat-to-beat annotations of data are converted to a T-second resolution (T is the duration of each data segment) by using a minimum percentage parameter P [18,20|: A data segment ‘classified asa true AF ony ifthe percentage of annotated AF beats Jn that data segment is more than P. A two-fold stratified cross- validation on the feature vectors (extracted from the dataset) is employed to train and test the classifier. At the testing phase ofeach fold, SVMs scares for the testing samples (distance of the testing samples to the separating hyperplane) are obtained and compared with a pre-set threshold ¢ to clasify corresponding testing samples, ‘The classification results ({rom each fold) is then compared to the ground truth (converted annotation) and number of True Positive (TP) False Negative (FN), True Negative (TN) and False Positive (FP) cases are calculated. Finally, the results from the two folds are ‘combined to calculate True Positive Rate (TPR) as TPR = TP/TP FN, False Positive Rate (FPR) as FPR = FP/FP-+TN and Accuracy (ACC) as ACC =TP-+TN/TP+-TN+EP+EN, In order to obtain the optimal values of the parameters of the proposed method (T. P and ¢), an exhaustive standard Receiver Operating Characteristic (ROC) Curve analysis is implemented as the following: Parameter T (length of the data segment in seconds) Js varied over a reasonable range of 10-120 with incremental steps of 5s, Parameter P (percentage threshold for annatation of AF data segments) is varied from 0% to 100% with incremental steps of 10%. Then for each specific value of T and P, an ROC curve 's derived by varying the score threshold ¢ from — 15 t0 15 with ‘incremental steps of 0.01. The ROC with the highest Area Under the Curve (AUC) is used to obtain the optimal values of parameters A x0 5 oo I: ttn “ie oso) ‘Tee sone) D gg g ie encont) Fig. 3. Examples of FCG sige dving four annua epsodes (A) Atal Sein; (ara ue; (© anetonal shyt: and (0) ater yen, 136 Asari e ob Computers in Blog and Medan 60 (2015) 152-142 T and P. Furthermore, the score threshold ¢ corresponding to the knee point af this ROC curve (closest point to ideal perfarmance af FPR=0 and TPR=1) is considered as the optimal value for score threshold. ‘To further investigate the efficacy of the chosen classification algorithm (SVM), we compare the ROC curve of AF detection using SVM classifier with those of three other popular classifers—viz, Naive Bayesian, Linear discriminant Analysis (LDA), and Logistic Regression. For this purpose, the ROC curves of other classifiers are ‘obtained similar to that of SVM by varying either the score thresh= ‘old (logistic regression), ora threshold on the AF posteriori (Naive Bayesian, LDA). Finally, we compare the performance of our proposed method with those of the existing AF detection methods in terms of various measures including accuracy, transition time, and computation time, 3. Results Fig, 4 presents typical ECG waveform prior to and during an AF (extracted from 6th AF episode of fle 4048 in MIF-BIM’ AFB database), along with its detail wavelet coefficients at levels 4-6 (dam, dein) and dem). This example demonstrates signature characteristics of an AF episode, namely irregular R-R interval and absence of P wave. It also shows the efficacy of wavelet transform to ‘capture the atrial activity, especially at detail coeficients. Fig 5 presents two examples of the application ofthe proposed ‘method (with T = 10 and P = 50%), on segments of ECG signal. The ECG signal of Fig. 5(A) contains the 8th AF episode of file 4936, while the ECG signal of Fig. 5(E) contains the 27th normal sinus rhythm (NSR) episode of fle 8219, From Fig, 5(D), we observe that by using a score threshold of ¢= 0, our proposed method is able to correctly identify the entire AF episode (a true positive case), Similarly, as displayed in Fig. 5(H), our proposed method is able to correctly identiy the entire non-AF episode (a true negative case). ‘The application ofthe proposed method resulted in few cases of false negative and false positive in detection of AF episodes. Fig. 6 presents two examples of those cases (a false negative and a false Positive case). The ECG signal of Fs. (A) contains the 2th AF episode Of file 4746, while the ECG signa of Fig. 6(E) contains the 26th NSR episode of file 4936. As Fig. 6(D) illustrates, wth T=10s, our proposed ‘method has missed the detection of the short AF episode (a false negative case). Note that in both ofthe two 10-s data segments which Partially contain the AF episode (corresponding to time between 30 and 50), only ess than P= 507% ofthe segment holds the AF pattern. ‘This could have contributed to missing the detection ofthe elated AF episode. On the other hand, Fg. 6(H) shows that although our proposed method has been able to identify the NSR episode, but its estimated duration is less than the duration ofthe annotated episode. ‘This is due tothe fact that calculated SVM scare for the dth 10's data segment (corresponding to time between 30 and 40) s slighty above the threshold ¢=0, and hence, this data segment has been falsely detected as AF (false positive case) Fig. 7 presents two more examples of false positive cases. The ECG signal of Fig. 7A) (extracted from file 6426) contains an atrial flutter episode with duration of approximately 118. As Fig. 7 (D) illustrates, few data segments along the corresponding atrial flutter episode have been wrongly detected as atrial fibrillation. The ECG signal of Fig. 7(£) contains a junction rhythm episode of fle 7879, Similarly (as Fig. 7(H) demonstrates), the proposed method has falsely labeled the junction shythm episode as an AF episode. In both examples, the itegularity associated with a non-AF episode (i, atrial utter, and junctional rhythm) has resulted in a false Positive case. To evaluate the prevalence of the false positive (and false negative cases), we employed the ROC curve analysis (Fig 8) ‘Note that although Fig. $ depicts the results of parameter value ‘optimization only for few values of T and P, we performed the a> C a, ay . 7 5 20 cs 30 > emienaryannanaeyanyaniae anil Asal wine Fig 4 An example of ECG signal pir to and dung AF (exacted fom 5 AF epsde om fle 4048 of MI-BE ARI database} an Us etal wavelet cet (a) vin sgna (8) deta cee at ee (C) deta cetent at evel 5; and (D} deta cotient a eve 6. S. Agar t lf Computers in ily and Meine 6 (205) 132-142 w a 2 : a i na ocd : = i oni : ond za T 2 i ' ew Lt | Pee, PU ae eee Time tel Tee wee | 3 oa * is § | Seo 4 Ta gt eae 8 ee ee Be? Gates eo Pe: Two ean of plc fh ep ta pare an 5 ta get cong 28 Fed (eg pa) t Tica eae lc at ctr Steen (tv sowa a he sae eo oa SE ‘Sree ons a tc nan Co eee eet 3M cote ot Seton oa ns) A e 5 l 5 5 rae Time) B F i g 2 | a a Sowa E nora il Ppa ee ee ‘rae ‘ime ce c G 2 | e 3 oo “ie et) oD H i i po 5 pon a a ) (amd Tc Fig 6. Samples of fe negative ad ale postive cast in AF detection using the proposed meth: et side pots ae related oan ECG data segment containing an AF ‘pode (A) Giga gal (@) ue annescon: (c)celated score output ot SVN Stes) ahd (0) dtecton reel AF versus nowAB) Right sie pot sees a csr coting aA epoe (0) rigor mnt (ee eo of Master 8 etn ea AF eo Asari e ob Computers in Blog and Medan 60 (2015) 152-142 A : , ; ; 3 a f*s25 "= 72 B F 4 g : j Frcs i i- im a Tomo c G 3 i : Me D 4 g ia in hb a Tinoisocora) ig 7. Examples of le postive cases in AF detection wing the proposed metho: Left ide plots ar related toa CG data Segent containing an ail fiter episode (a) ovina (8) te annotatn,(C)caeulated are utp or SVM cassie) and (D detection Res (AF Versus aba-AP) Right ide pre teated tan FCG aa Segment containing an epaade with juan (E) gn signa (Fue annotation c)cauated sar (ouput o SVM caster) and (3) tection esl (AF versus non-A Timo(ocond) A, B TPR ena) wa ewmnay) FPR tapes) FPR isp) ig 8 Results of parameter vale optimization using ROC curve alysis (A) P= 07 varying score these varying T 2 15,30 48 and 6s: and) =30 varying sore hresold varying Pa HX, 308,50, 708 ad 90. (A) and all 5 ROC curves of Fig. 8(8) are above 0.992. Ths indicates that our proposed method is able to maintain its high accuracy actual parameter optimization on the entire range of each para meter, as described in Section 2.3. t can be noticed that the ROC curves of both subplots (Fis. (4.8) are very closely spaced. In fact, ‘our results showed that the AUC of all four ROC curves of Fis. § regardless of the choice of the parameter values T and P. Never- theless, the highest AUC of 0995 is obtained when T=30 and 5. Agar tl / Computers in ily and Meine 6 (201) 132-142 19 A inate a £ FPR tsp) By “Te ens) FPR (1 seotaty) ig. 8. Comparison of AF detection resus tens of RC curves (A) Using vais easton grt: and (B) proposed method versus ta of Weng method 25) Result ofthe performance of eur proposed method on MT-BI ARB databise (cve-fold soa cras-valiation sing the optimal aes of te parameters Speciciny(®) curacy) » a0 on om point to the ideal performance of FPR: = 005, Fig. 9(A) presents results of AF detection (with T=30 and ‘P= 50%), obtained from applying various classification algorithms ‘on the extracted 28-dimensional feature space. Note that similar to Fig, 8 all the ROC curves are calculated by using. a two-fold stratified cross validation as described in Section 2.3, We observe that Naive Bayesian classifier with AUC of 0.915 has the lowest accuracy. LDA and Logistic regression perform slightly better with AUC of 0.837 and 0.949, respectively, while our proposed SVM. lassifier achieves the best performance with AUC of 0.996. Fig 9B) compares the result of AF detection using our proposed method with that of Weng et al which have used a similar SWT approach [25] We abserve that AUC of Weng method is 089, and hence our proposed method outperforms Weng method by 10% Furthermore, at sensitivity of 100% the specificity of the proposed ‘method i 86.3% while the specificity ofthe Weng method is only 2%. Finally, Table 1 summarizes the result of the performance of our proposed method on MIT_BIH AFB database (two-fold stratified cross- validation) using the optimal values of the parameters P=50% an $= 0.05, With optimal data length of T= 30, our method achieves sensitivity of 973, specificity of 971% and an accuracy of 971% To complete our study. we also analyzed the transition delay and computation time of our proposed method. Note that in all segment-based AF detection methods, the transition from a non AF to an AF episade is detected with a delay (Le. transition delay). For example, the transition delay in the Fig. 5 is 68s. In general, transition delay isa function of the length ofthe data segment (7, and a longer data segment will induce a larger transition delay. ‘Our analysis showed that the median transition delay of our proposed method for T= 10 is 98's ‘The computation time of our proposed method implemented Matlab 2012a (on an Intel® core™ CPU 3.5 GHz, 32 GB RAM, G4bit OS) is approximately 35 ms and 40 ms, with T=10 and T=30, respectively. Given this short computation me, our pro- posed algorithm is easily realizable in real-time for automatic AF detection, 4. Discussion ‘Automatic and accurate detection of AF using ambulatory ECG ‘monitoring could lead to earlier diagnosis, and consequently provides more opportunities for effective management or treat- ‘ment of the condition, and avoiding the complications associated with chronic AF ‘Over the last two decades, several algorithms have been devel- ‘oped for automatic AF detection. These algorithms ether rly on the absence of P-waves or R-R irregularities, or combination of both characteristics to detect AF episodes. Table 2 lists some of the most- cited and recently proposed AF detection algorithms. Among the listed methods, algorithm #1 isa P-wave-based method, algorithms 172-16 are R-R interval-based algorithms, and algorithms # 7-19 tse the combination of these characteristics to detect AF. Note that all ofthese algorithms require a pre-processing step where the P andjor R peaks of ECG signals need to be located. As a result, their performances depend on the accuracy of peak detection step and they do not perform well in the case of missing peaks or etro- neously detected peaks, ‘The current work introduces a novel AF detection method which does not require any P andjor R peak detection step. Our method leverages on the high time-frequency resolution of stationary wavelet transform, and captures atrial activity by calulating peal-to-average power spectrum ratio over diferent frequency bands. Furthermore, we extract the log-energy entropy ofthe wavelet coefficients, to enhance the performance of the method, The extracted features are then Classified into AF (and non-AF) cases using a support vector machine. 4.1, Number of parameters and the choice of parameters values ‘The proposed algorithm has only few parameters to optimize: segment length T, minimum AF percentage P, and SVM score threshold ¢. Interestingly, the results of our parameter optimization (Fig. 8) indicated that the method maintains its high accuracy regardless of the choice of T and P values (less than 03% change 0 S Asgare a (Computes in Bog Table ond Meine 6 (205) 132-142 Comparison athe performances recent AF detection lors on MF-BIY AFB database, The bol percentages aside the paenteses age terete r-eauaed by © Algoritn narme ‘Sensivity Speicy weakness points relate to the proposed method) oo 1 Slocum etal > (628) (75) Needs eat detection has lower performance 2 Tana 200 |is!” 644972. —_Neads beat etectian has lower performance has ager memory consumption (2) a 3 Sararetal isp 97588 Needs beat tection as large memory consumption. neds longer data segment suarante high perfomance 4 Dashetal 20)" $44 95:1 Nees eat detection, has ower perfomance, has larger ans ely hes Mgbe sensi tthe Selection of ameter values 5 Huonget at (267 961 981——_-Nende beat detection, as ager memary consumption. as Lge ranson delay & tmceeatiiiP 382 597 Nende brat Gctectan fas ges sey fo the slim of parameter eles rer (56) 7) 1 fatoenagon eeak 2 (055) Neste best easton hae lower performance rar eo © fanpecal24r 982° 975 Neds beat detection, has Larger memory consumption, needs longer data Segment to guarantee high performance, has ewer AF episode detection rae 10 proposed metnod $7071 © Bcc on he abience of ova Bad on RoR equa. “Based on combination of -R equity and absence of Pave in AUC, In contrast to our method, many ofthe existing algorithms ned to tune atleast 5 parameters to guarantee a high performance [19-21], Moreover, the accuracy of these methods is much more sensitive t0 the choice of those parameters. For example, in Dash [20] changing the segment length from 128 beats to G4 beats

    You might also like