We are searching data for your request:
Upon completion, a link will appear to access the found materials.
I saw this youtube video on ECG readings and cardiac dipole vector locus of the heart across the three leads. Posting screenshots from it:
My question is how to obtain the graph (2nd image) from the 3 graphs in the first image? Should I simply add them or is there something more to it?
Chapter 4 - Machine learning in biomedical signal processing with ECG applications
The research in the field of ECG signal analysis and classification is an active research topic. ECG signal analysis plays an essential role in detecting cardiovascular diseases (e.g., occlusion of coronary arteries, heart enlargement, conduction defects, rhythm, and ionic effects). Cardiologists are trained on how to interpret ECG signals for identifying arrhythmias. For example, ECG beats are examined by determining distinctive morphological and interval-based features. However, manual analysis is time-consuming, requires expert training, and is prone to error.
Automated ECG signal analysis technology has become widely available thanks to the advances in machine learning and biomedical signal processing. ECG analysis is noninvasive and has shown impact in various applications including medicine, emotion recognition, biometric identification, and sports wearable technology. This chapter is concerned with automated electrocardiogram (ECG) analysis for the diagnosis of cardiovascular diseases. The main aim of this chapter is to help the biomedical engineer to build a machine learning model to perform automatic classification of ECG beats. I will start with an overview of clinical ECG, ECG views, heartbeat types, and arrhythmias. Then a simple model for ECG heartbeat classification is illustrated, showing the various stages from data selection, preprocessing, feature generation, and training a machine learning model.
Get full journal access for 1 year
All prices are NET prices.
VAT will be added later in the checkout.
Tax calculation will be finalised during checkout.
Get time limited or full article access on ReadCube.
All prices are NET prices.
This section covers following aspects: In Data Acquisition, we describe in detail the data acquisition process and in Preprocessing we discuss the applied preprocessing steps in order to facilitate a widespread use for training and evaluating machine learning algorithms.
The raw data acquisition was carried out as follows:
The waveform data was automatically trimmed to 10 seconds segments and stored in a proprietary compressed format. For all signals, we provide the standard set of 12 leads ( I,II,III,aVL,aVR,aVF,V1–V6 ) with reference electrodes on the right arm. The original sampling frequency was 400 Hz.
The corresponding metadata was entered into a database by a nurse.
Each record was annotated as follows:
An initial ECG report string was generated by either:
67.13% manual interpretation by a human cardiologist
31.2% automatic interpretation by ECG-device
4.45% validation by a human cardiologist
26.75% incomplete information on human validation
1.67% no initial ECG report.
In Quality Assessment for Annotation Data (ECG Statements), we provide a more extensive discussion on this step.
The report string was converted into a standardized set of SCP-ECG statements including likelihood information for diagnostic statements.
The heart’s axis and the infarction stadium (if applicable) was extracted from the report.
A potential second validation (for first evaluation in case of a missing initial report string) was carried out by a second independent cardiologist, who was able to make changes to the ECG statements and the likelihood information directly. In most cases, the deviating opinion was also reported in a second report string.
Finally, all records underwent another manual annotation process by a technical expert focusing mainly on qualitative signal characteristics.
The waveform files were converted from the original proprietary format into a binary format with 16 bit precision at a resolution of 1 μV/LSB. The signals underwent minor processing to remove spikes from switch - on and switch- off processes of the devices, which were found at the beginning and the end of some recordings, and were upsampled to 500 Hz by resampling. For the user’s convenience, we also release a downsampled version of the waveform data at a sampling frequency of 100 Hz.
With the acquisition of the original database from Schiller AG, the full usage rights were transferred to the PTB. The Institutional Ethics Committee approved the publication of the anonymous data in an open-access database (PTB-2020-1). ECGs and patients are identified by unique identifiers. Instead of date of birth we report the age of the patient in years at the time of data collection as calculated using the ECG date. For patients with ECGs taken at an age of 90 or older, age is set to 300 years to comply with Health Insurance Portability and Accountability Act (HIPAA) standards. All ECG dates were shifted by a random offset for each patient while preserving time differences between multiple recordings. The names of validating cardiologists and nurses and recording site (hospital etc.) of the recording were pseudonymized and replaced by unique identifiers. The original data contained implausible height values for some patients. We decided to remove the height values for patients where the body-mass-index calculated from height and weight was larger than 40.
The ECG data was annotated using a codebook (SCP-ECG v0.4 (Annex B)) of ECG statements that preceded the current SCP-ECG standard 12 . All annotations were converted into SCP-ECG statements by accounting for the minor modifications that occurred between the release of the codebook and the publication of the final standard.
3.1 Denoising of ECG signal
The discrete wavelet transform (DWT) method is used for soft-thresholding denoising [ 15 ]. We use the db3 wavelet to process the original ECG signal. To calibrate the ECG signal baseline shift, the normalised least mean square algorithm is used to conduct adaptive noise filtering. The results show that this method can effectively correct the baseline shift and noise, whilst maintaining the geometric characteristics of the ECG signal. Fig. 4 shows the comparison of the original and denoised signals.
Comparison of the original ECG signal and the denoised ECG signal-based DWT
3.2 Heartbeat segmentation
We segment the ECG data into the structure of the ECG beat. First, we use the effective R wave annotation from the MIT-BIH database to find the R wave position, and to calculate each RR interval length. Second, we segment 40% of the RnRn-1 interval and 60% of the RnRn + 1 interval on the every R wave position. Finally, we resample the ECG beats using the mean length of the RR interval, to obtain the single ECG beat in the same length. The result of the ECG segmentation is shown in Fig. 5.
ECG signal segmentation
3.3 Dimension reduction
In signal processing, ICA [ 16 ] is a computational method for separating a multivariate signal into additive subcomponents. This is done by assuming that the subcomponents are non-Gaussian signals, and are statistically independent from each other. Compared to principal component analysis (PCA), the ICA method not only realises decorrelation, but also accounts for high-order statistical independency. The ICA method has been widely used in the fields of biomedical signal processing, speech processing, and communication [ 17-19 ]. Moreover, as mentioned in [ 20 ], the useful ECG data at times give better a feature selection by using the ICA method.We use a statistical ‘latent variables’ model to introduce the ICA method. Assuming that we can observe linear mixtures of n independent components (ICs) (1) (2) (3) In the ICA model, we assume that the components are statistically independent, and that the ICs must have non-Gaussian distributions. The output of the ICA method is an estimate of un-mixing matrix , such that is (4)
We can observe that, , so is the generalised inverse matrix of A. The estimated ICs will be a mixture of those true independent sources with element of as a scale factor. To reduce the database dimension, we use A, which represents the coefficients, as our clustering algorithm input. We observe that only 20 ICs can make the correlation coefficient, between the reconstructed signal and the original signal, reach 93.1%, as shown in Table 2. Although a higher number of ICs result in better representation of the ECG beat, the correlation coefficient does not increase significantly after 20. Therefore, we set the desired number of ICs in our experiment to 20. Fig. 6 shows all 20 ICs from the ECG beats presented in Fig. 5.
ICs of the ECG beats, the desired number of ICs is 20
|Number of ICs||Correlation coefficient|
How to obtain normalised ECG given ECG readings across lead 1, 2, and 3? - Biology
Objective To evaluate the diagnostic performance of a deep learning system for automated detection of atrial fibrillation (AF) in photoplethysmographic (PPG) pulse waveforms.
Methods We trained a deep convolutional neural network (DCNN) to detect AF in 17 s PPG waveforms using a training data set of 149 048 PPG waveforms constructed from several publicly available PPG databases. The DCNN was validated using an independent test data set of 3039 smartphone-acquired PPG waveforms from adults at high risk of AF at a general outpatient clinic against ECG tracings reviewed by two cardiologists. Six established AF detectors based on handcrafted features were evaluated on the same test data set for performance comparison.
Results In the validation data set (3039 PPG waveforms) consisting of three sequential PPG waveforms from 1013 participants (mean (SD) age, 68.4 (12.2) years 46.8% men), the prevalence of AF was 2.8%. The area under the receiver operating characteristic curve (AUC) of the DCNN for AF detection was 0.997 (95% CI 0.996 to 0.999) and was significantly higher than all the other AF detectors (AUC range: 0.924–0.985). The sensitivity of the DCNN was 95.2% (95% CI 88.3% to 98.7%), specificity was 99.0% (95% CI 98.6% to 99.3%), positive predictive value (PPV) was 72.7% (95% CI 65.1% to 79.3%) and negative predictive value (NPV) was 99.9% (95% CI 99.7% to 100%) using a single 17 s PPG waveform. Using the three sequential PPG waveforms in combination (<1 min in total), the sensitivity was 100.0% (95% CI 87.7% to 100%), specificity was 99.6% (95% CI 99.0% to 99.9%), PPV was 87.5% (95% CI 72.5% to 94.9%) and NPV was 100% (95% CI 99.4% to 100%).
Conclusions In this evaluation of PPG waveforms from adults screened for AF in a real-world primary care setting, the DCNN had high sensitivity, specificity, PPV and NPV for detecting AF, outperforming other state-of-the-art methods based on handcrafted features.
In this section, we review forecasting methods for temporal data particularly with applications to healthcare domain. These forecasting methods can be divided into two main categories: (i) traditional machine learning-based methods and (ii) deep learning-based methods.
For traditional machine learning-based forecasting methods, two representative approaches are support vector machine (SVM) and artificial neural network (ANN). Wu et al.  employed SVM to predict heart failure more than six months via vast electronic health records (EHR). The highest value of area under curve (AUC) for SVM is around 0.75. Santillana et al.  utilized the SVM to forecast estimates of influenza activity in America. Yu et al.  used the SVM to predict one-day-forward wellness conditions for elderly and achieved the forecasting accuracy of around 60%. Meanwhile, the ANN also obtained widely application in health care domain. Suryadevara et al.  took advantage of the ANN to forecast the behavior and wellness of elderly and deployed it into a healthcare prototype system. Srinivas et al.  employed the ANN to predict heart diseases like chest pain, stroke and heart attack. The prediction performances of these traditional machine learning-based methods are difficult to meet the precisely forecasting demands of elderly. So, researchers shifted their attention to cutting-edge deep learning-based forecasting methods.
In recent years, deep learning-based methods like recurrent neural network (RNN) has been achieved a big success in natural language processing, speech recognition, and machine translation [28–31]. Researchers also attempted to solve the problems in healthcare domain using these cutting-edge approaches [32–34]. Ma et al.  proposed an end-to-end simple recurrent neural network to model the temporality and high dimensionality of sequential EHR data to predict patients’ future health information. The experimental results based on two real world EHR datasets showed that their model improved the prediction accuracy significantly. Choi et al.  explored recurrent neural network whether improving initial diagnosis of heart failure compared to traditional machine learning-based approaches. Experimental results proved that recurrent neural network could leverage the temporal relations and improved the prediction performance of incident heart failure. Choi et al.  also proposed an interpretable forecasting model based on recurrent neural network. This deep model was tested on a large EHR dataset and demonstrated its superior prediction performance. Therefore, two popular deep learning-based approaches called long short-term memory network (LSTM) [35, 36] and bidirectional long short-term memory network (BiLSTM)  are utilized to forecast one-day-forward wellness conditions for elderly in this study. Meanwhile, two traditional machine learning-based methods of SVM and ANN are also employed for model selection.
Why is it done?
An ECG gives two major kinds of information. First, by measuring time intervals on the ECG, a doctor can determine how long the electrical wave takes to pass through the heart. Finding out how long a wave takes to travel from one part of the heart to the next shows if the electrical activity is normal or slow, fast or irregular. Second, by measuring the amount of electrical activity passing through the heart muscle, a cardiologist may be able to find out if parts of the heart are too large or are overworked.
Four major issues must be highlighted regarding the adequacy of the studies conducted so far. First, while a great effort has been spent in feature selection and classifier design, it is not yet clear what is the best set of features and classification scheme for ECG biometrics (hierarchical, ensemble etc.). Non-fiducial based techniques can reduce the computational effort as well as the error rate due to the ECG waves recognition. Therefore, it is expected that the new techniques to be developed will use fiducials and non-fiducial based features in order to catch the best of both approaches. Further analysis should be addressed on the use of single lead recordings and the study of features which are not dependent on the recording sites (e.g. fingers, hand palms).
Second, as regards the population size, the majority of the studies have been conducted on a small population (about a few tens of subjects). Therefore, the applicability of ECG biometric recognition on a large scale (real life authentication scenario) it is not yet proven.
Third, almost all studies (except for  and  ) ignored the variability of the ECG during life span (i.e. variability induced by work, ageing, iterate sport activity etc.) moreover, only few studies [57, 83, 136] considered the applicability of these techniques when subjects suffer from pathological conditions. ECG recognition in pathological subjects is another aspect worth of additional investigations.
Fourth, it must be emphasised that, while guidelines are available for ECG acquisition in the clinical scenario, there is still a lack of standardisation on ECG acquisition (number of leads and their positioning, sampling frequency, number of bits, filtering, type of electrodes, number of leads etc.) for biometrics applications. However, ECG databases for biometric recognition should ideally include recordings, at a given sampling frequency and conditions, from the same subjects in different circumstances (e.g. relaxed, during and after physical training) and along a period of several years.
If addressed, the mentioned challenges will contribute to move this promising technique from its state of adolescence to a proper daily life adoption.
Background and Purpose—An M-shaped bifid notch on the ascending branch, or on the zenith, of the R wave in inferior ECG leads (II, III, aVF), so called “crochetage,” is an indicator of ostium secundum atrial septal defects. The pathophysiology underlying this finding remains unknown. A crochetage pattern has not been previously reported in patients with patent foramen ovale (PFO) however, the location of this defect and the secundum atrial septum are similar. The purpose of this study was to determine the prevalence of crochetage in cryptogenic stroke patients with or without PFO.
Methods—A conservative selection scheme was used to identify patients likely to have had PFO-associated strokes (ie, cryptogenic) and to exclude any structural, functional, or vascular heart disease responsible for ECG changes. All patients had a standard 12-lead ECG. The prevalence of crochetage in each group was determined.
Results—Sixty consecutive patients were studied (28 with echo-documented PFO and 32 echo-negative control subjects). The crochetage pattern was present in at least 1 inferior limb lead in 10 of 28 PFO patients (36%) and 3 of 32 control subjects (9%) (P<0.05). The sensitivity and specificity of the crochetage pattern for diagnosis of PFO in cryptogenic stroke cases were 36% and 91%, respectively positive predictive value was 77%.
Conclusions—The finding of an ECG crochetage pattern may help to identify stroke patients with PFO, may help to streamline their diagnostic workup, and may warrant future studies to determine its value in stratifying stroke risk in patients with PFO.
Paradoxical embolus through a PFO, a potential channel between the atria, has recently been proposed as a major cause of otherwise cryptogenic embolic stroke. 1 The primary, but indirect, evidence rests on the significantly higher prevalence of echocardiographic diagnosis of PFO, especially in young stroke patients without other known causes. 1 2 3 4 Quick and accurate diagnosis of PFO is important in patients with stroke or TIA to prevent early cerebral or systemic embolic recurrences. However, in common practice, diagnosis of PFO is usually delayed because patients are scheduled for echocardiography days after the onset of stroke. Furthermore, in most centers, transthoracic color Doppler echocardiography is the choice in routine evaluation of stroke patients, but its yield in detecting PFO is very low. 5 Moreover, the image quality of color TTE is often degraded during the Valsalva maneuver, a maneuver crucial for creating a right-to-left shunt in those persons without spontaneous shunting. More sensitive, albeit more invasive, techniques such as transthoracic contrast echocardiography, transesophageal contrast echocardiography, and transcranial contrast Doppler ultrasonography are required for the diagnosis of PFO. A readily available indicator of the presence of PFO in stroke patients could streamline the diagnostic evaluation and patient management.
A notched pattern of the R wave, so called “crochetage,” in the inferior limb leads has recently been demonstrated to be associated with ostium secundum-type atrial septal defect (ASD). 6 The exact mechanism leading to a crochetage pattern in ASD is not known. Crochetage has not been reported previously in patients with PFO. However, the similar location of PFO and ostium secundum ASD, and the hemodynamic similarities between a large PFO and an ASD, motivated us to investigate the prevalence of crochetage in cryptogenic stroke patients with PFO.
Subjects and Methods
We examined the hospital records of patients admitted between March 1990 and March 1997 with the diagnosis of first-ever ischemic stroke or TIA. The procedures followed were in accordance with institutional guidelines and with the approval of the Institutional Review Board. All patients had clinical symptoms consistent with a specific arterial distribution in the retina, cerebral hemisphere, or brain stem. Symptoms were transient and lasted less than 24 hours in patients with TIA, or longer than 24 hours in those with ischemic stroke. All patients had brain CT or MRI studies compatible with their diagnoses. All patients underwent routine laboratory studies (blood chemistries, cell counts), 12 lead ECG, TTE and/or TEE, and noninvasive vascular studies that included duplex carotid Doppler ultrasonography and/or transcranial Doppler sonography. Twenty-four-hour Holter monitoring, conventional cerebral angiography and/or magnetic resonance angiography, and blood tests for hypercoagulable states or immunologic abnormalities were performed only in selected cases in whom no other cause of stroke could be identified.
Our selection process sought to identify a patient cohort most likely to represent PFO-associated strokes and a control group of patients with no identifiable cause of stroke (in other words, a study group of patients with cryptogenic strokes with PFO, and a control group of patients with cryptogenic strokes without PFO). Excluded from our study were: patients with any degree of stenosis or occlusion of a major extracranial or intracranial vessel ipsilateral to the symptomatic side, not only those stenoses that may have caused hemodynamic abnormality but also those that may have served as a source of emboli patients with small infarctions (less than 15 mm in diameter) in the territory of perforating arteries either associated with 1 of the 4 classic lacunar syndromes (pure motor hemiparesis, pure sensory stroke, ataxic hemiparesis, and sensory-motor stroke) or with risk factors for small vessel disease such as diabetes mellitus and hypertension patients with other rare causes of stroke such as vasculitis, arterial dissection, or complicated migraine and patients with any structural, functional, or vascular heart disease that might produce ECG changes or that may serve as a source of embolus. In accordance with the latter criterion, we excluded all patients with any history of clinical heart disease, with any ECG abnormality (myocardial ischemia, infarction, atrioventricular or intraventricular conduction block, arrhythmia, pericarditis), or with any echocardiography (ECHO)–documented cardiac pathology (wall motion abnormality, cardiomyopathy, pericardial effusion or tamponade, segmental left ventricular hypertrophy, ASD, ventricular septal defect, atrial septal aneurysm, or heart valve disease with the exception of mitral valve prolapse.
All patients had a standard 12-lead ECG with a sensitivity of 10 mm/mV and paper speed of 25 mm/s. Crochetage pattern was described as an M-shaped notch on the ascending branch, or at the top, of the R wave in inferior limb leads (II, III, and aVF) (Figures 1 and 2 ) the notching must be persistent in all QRS complexes in an individual lead in a given tracing, or—in the case of multiple tracings—across the various studies. All ECG traces were analyzed with respect to the absence or presence of the crochetage pattern and the number of leads that exhibited notching. Analysis was performed by 2 examiners who were blind to the study groups. Contrast echocardiography studies were performed by injection of 7 mL of saline agitated with 1.0 mL of air into an antecubital vein at rest and with Valsalva maneuver. A PFO was diagnosed if at least 3 microbubbles were seen in the left atrium within 3 cardiac cycles after maximum opacification of the right atrium.
We compared cerebral ischemic lesion size in PFO patients with or without crochetage by assuming that the size of the embolus would be greater in larger infarctions, ie, involving cortical and subcortical territories of a major intracerebral artery. This type of distribution of infarction referred to the stem or main branch occlusions of the anterior, middle, and posterior cerebral arteries. Infarctions isolated either to cortex or subcortical structures, brain stem, or cerebellum were assumed to be small.
All data were expressed as mean±SD. Frequency data were given as percentage and the significance was assessed by χ 2 . This was replaced by Fisher‘s exact test when a cell frequency was less than 5. Results were considered significant at P<0.05. Agreement between the examiners for identifying a crochetage was evaluated using the κ statistic. 7 A κ value of 1 indicates perfect agreement, whereas 0 indicates only chance agreement. In general, excellent agreement refers to values >0.81, 0.61 to 0.80 indicates good agreement, and values <0.20 indicate poor agreement.
Among a total of 1470 patients with first-ever stroke or TIA, there were 167 patients with cryptogenic strokes of these, 60 cases fulfilled our conservative eligibility criteria, and comprised 28 cryptogenic stroke patients with PFO and a control group of 32 cryptogenic stroke patients without PFO.
Clinical features of the study patients are summarized in the Table . The mean age was lower in patients with PFO (45.0 versus 52.1 years). The male-female ratio and the clinical type of ischemic attack between groups were not significantly different. The mean number of cardiovascular risk factors (including hypertension, hyperlipidemia, obesity, smoking, and diabetes mellitus) was lower in the PFO group (0.5±0.7 versus 1.0±0.9). There was a difference between the number of TTEs and TEEs performed in each group TEE examination was performed in 11 patients in the PFO group and in 10 patients in the control group. Moreover, low yield color TTE alone, rather than TTE study with contrast injection, was performed more often in the control group (1 versus 9 patients). Thus, contrast TTE and/or TEE studies were obtained in 96% of cases (27 patients) in the PFO group, whereas in only 72% (23 patients) of patients in the control group (P<0.05). A minimal degree of mitral valve prolapse was present in 3 patients in the PFO group and in 1 of the control subjects. Deep venous thrombosis was detected in 4 patients, all in the PFO group. Four patients, 3 with deep venous thrombosis, eventually underwent surgical closure of the PFO.
For each patient, at least 1 ECG was accessible from the cardiac database unit 2 or more tracings were available in 60% of PFO patients and in 44% of the control patients. In 3 cases, ECGs were obtained before the stroke (2 in the PFO group and 1 in the control group). The time between stroke and the nearest ECG tracing varied from 1 day to 7 months, but it was less than 2 days in most instances. Examiner 1 (S.A.A.) determined crochetage pattern in at least 1 inferior limb lead in 10 of 28 patients in the PFO group and in 3 of 32 control patients. Examiner 2 (F.S.B.) rated a crochetage in 11 patients in the PFO group and in 2 patients in the control group. Concordance among the 2 examiners regarding the presence of a crochetage was 90%. After adjustments were made based on interobserver agreement (crochetage in 10 patients in the PFO group versus 3 patients in the control group), the difference between the groups with respect to the presence of crochetage was statistically significant (P<0.05) (Figure 2 ). The sensitivity and specificity of crochetage for the diagnosis of PFO in cryptogenic stroke patients were found to be 36% and 91%, respectively. The positive predictive value was 77%, and the negative predictive value was 62%. The difference in prevalence of crochetage remained significant (P<0.05), even after the exclusion of patients in each group evaluated only by color TTE (sensitivity, specificity, positive predictive value, and negative predictive value were 37%, 91%, 83%, and 62%, respectively), or after exclusion of the patients with mitral valve prolapse, ie, 3 patients in the PFO group (1 with crochetage) and 1 patient in the control group (who did not have crochetage) (36% sensitivity, 90% specificity, 75% positive predictive value).
In the PFO group, crochetage was noted in 9 patients in only 1 lead, and in 1 patient in 2 leads. Crochetage was present in 6 patients in lead III, in 5 patients in aVF, and in 0 in lead II. In the control group, all 3 patients with crochetage had it in only 1 lead. As defined in the “Materials and Methods,” crochetage was a consistent finding from 1 ECG to the next. A PFO had been ruled out in these 3 patients of the control group (by a TEE in 1, a contrast TTE in 1, and only by a color TTE in 1).
Large (ie, cortical-subcortical) cerebral infarction occurred in 60% of PFO patients with crochetage (6 of 10 cases) but in only 39% (7 of 18) of PFO patients without crochetage. In contrast, small cerebral lesions isolated either to cortical or to subcortical structures, or to the brain stem or cerebellum, tended to be more frequent in PFO patients without crochetage (9 versus 2 cases) however, this difference did not achieve statistical significance (P=0.15).
Although not a primary aim of the present study, we also determined the frequency of incomplete right bundle branch block pattern (incomplete RBBB) (R′ or r′ in lead V1 or V2 and R′ greater than R in V1 and V2 and QRS duration less than 120 milliseconds, or R peak time >50 milliseconds in lead V1 or V2 when QRS duration was < 120 milliseconds). There were 4 patients showing incomplete RBBB both in the PFO group and in the control group. Three of 4 patients with the incomplete RBBB pattern in the PFO group, but only 1 in the control group, exhibited the crochetage pattern.
The foramen ovale is a channel between the atria that enables passage of blood from the inferior vena cava into the left atrium in fetal life. After birth, pressure changes between the pulmonary and systemic circulations can seal the opening by keeping the valve of the foramen ovale opposed to the ostium secundum septum. However, this is not always the case autopsy studies demonstrate patency in as many as 35% of adults. 8 9 10 A PFO has the potential to permit passage of emboli from the venous into the arterial circulation. 11 12 13 14 15 Lechat et al 1 demonstrated an association between PFO and cryptogenic strokes in patients less than 55 years old. The prevalence of PFO was 24% in patients with an identifiable cause for stroke, 40% in patients with no identifiable cause but risk factors, and 54% in patients without identifiable cause or stroke risk factors. Other studies also confirmed a similar association between PFO and otherwise cryptogenic ischemic stroke. 16 17 18 In the present study, our strict inclusion criteria markedly reduced the sample size but minimized contamination of the population with patients who had stroke unrelated to PFO or ECG changes based on other cardiac disease. We excluded all patients with either known or potential cardiac disease.
The actual role of PFO and the variables that determine its role in paradoxical embolism are still not well understood. In addition to risk factors for clotting in the pelvic and leg veins, major determinants might include the size of defect, the degree of right-to-left shunting, direction of current flow in the right heart, range of right-sided heart pressures, and the variable degrees of closure that the valve makes during different periods of the cardiac cycle. Moreover, for paradoxical embolism to occur a thrombus in the venous circulation must enter the right atrium and be directed through the foramen while it is open. Many of the parameters that determine passage of thrombus mentioned above are difficult to measure. Currently, the most practical and sensitive diagnostic method is transesophageal contrast echocardiography, which can show the presence of a PFO with approximately 80% sensitivity. 5 19 20 Transcranial contrast Doppler sonography is also sensitive in detecting PFO, comparable to that of TEE. 21 22 23 However, an echocardiographically documented PFO may be incidental rather than a causative finding. Determining more specific echocardiographic, ECG, and deep venous system characteristics for paradoxical embolus as the cause of stroke would aid in the clinical decision of whether to anticoagulate or to close the PFO.
The ECG pattern of incomplete RBBB has been known as a marker of ASD for at least 40 years 24–26 it has been postulated to occur due to selective hypertrophy of the basal portion of the right ventricle or to stretching of the peripheral conduction fibers. 27 28 29 30 Another ECG pattern, independent of incomplete RBBB, in ASD is crochetage: an early M-shaped notch on the R wave of the QRS complex in the inferior limb leads. 31 Crochetage, when present in only 1 lead, has a sensitivity of 73.1%, a specificity of 92.6%, and a positive predictive value of 69% for the diagnosis of ostium secundum ASD, and achieves a specificity of 100% if present in all 3 inferior leads. 6 Heretofore, the pattern has not been associated with any other cardiac conditions, and the pathophysiology is not known however, it has been reported to disappear from 1 or more leads after surgical closure of the ASD. 6 To the best of our knowledge, no specific ECG pattern has been associated with PFO prior to the current report. Here, we demonstrate a statistically significant increase in the prevalence of a crochetage pattern in the inferior ECG limb leads in patients with PFO and cryptogenic stroke as compared with control patients with cryptogenic stroke without demonstrable PFO. Two blinded examiners detected crochetage in at least 1 inferior ECG lead in 36% of PFO patients as opposed to only 9% of control patients. The low sensitivity suggests that a routine ECG would not be a useful screening test for PFO. However, ECG is an almost uniformly available clinical evaluation tool in all patients with stroke or TIA, principally to rule out other cardiac abnormalities that may serve as potential sources of emboli. Given the high specificity (91%) and moderately high positive predictive value (77%), recognizing a crochetage pattern may increase the clinical suspicion of paradoxical embolism. It may be helpful in streamlining the diagnostic evaluation, especially in a young, otherwise healthy patient with TIA or stroke for example, within minutes of a patient’s evaluation in the emergency ward, a certain degree of suspicion of PFO-related stroke can be generated, a TEE can be requested with alacrity, and a search for the source of the embolus is initiated with lower extremity ultrasound studies (and magnetic resonance or contrast venography, if necessary). Detection of crochetage does not preclude an echocardiographic study. On the contrary, it may accelerate the clinical arrangements to obtain early echocardiography with techniques more sensitive for PFO (contrast TTE or TEE). Our results may also be helpful in alerting the physician to perform a bedside transcranial Doppler sonography study with contrast injection, 22 and–if applicable–to take additional precautions, such as filtering all intravenous lines or initiating early anticoagulation.
Heller et al 6 reported that the presence of crochetage, and the number of leads exhibiting it, correlated both with the degree of left-to-right shunting and with the size of the ASD. It had been previously shown 32 33 that both the degree of right-to-left shunting and size of the PFO are larger in patients with arterial ischemic events. Our data showed a trend toward larger infarct size in PFO patients with crochetage than in PFO patients without crochetage. Three of 4 patients who were finally referred to surgery for closure of PFO exhibited crochetage. The reason for closure was coexisting fresh deep venous thrombosis in 3 patients and recurrent cerebral embolism with multiple infarctions in 1 patient. Unlike the reports from ASD studies, the crochetage pattern remained unchanged after the closure in each of the 3 patients.
The current study was limited by the relatively small sample size that resulted from very conservative selection criteria. Since patients with any ECG abnormality or a known cardiac disease were excluded, the impact of cardiac conditions on the ECG crochetage pattern remains to be studied. Another limitation of this retrospective study is the diverse methods of investigation used for the diagnosis of PFO. More than 20% of patients in the control group were evaluated only by color TTE. Because of the relatively lower sensitivity of this technique, PFO might have been missed in some cases. However, we suspect that 2 of 3 patients with crochetage in the control group may have had PFO since they did not have a TEE study. It is difficult to arrive at the true predictive value of the crochetage pattern for stroke due to paradoxical embolus without definitive knowledge about the incidence of other causes of stroke in the patients with PFO, specifically in those with and without crochetage.
In conclusion, the finding of a crochetage pattern may serve as a readily available ECG marker to motivate the search for PFO or ASD in patients with stroke or TIA. This study was performed in patients without heart disease or stroke risk factors other than PFO. Future prospective studies are needed to establish the relation of crochetage to PFO in the general population. It will be especially important to determine whether the presence or absence of the crochetage pattern correlates with stroke risk in persons with PFO. The clinically significant hypothesis raised by this study is whether the degree of shunting in patients with PFO correlates with the presence of crochetage, as it does in patients with secundum-type ASD.