- Research
- Open Access
Automatic classification of long-term ambulatory ECG records according to type of ischemic heart disease
- Aleš Smrdel^{1}Email author and
- Franc Jager^{1}Email author
https://doi.org/10.1186/1475-925X-10-107
© Smrdel and Jager; licensee BioMed Central Ltd. 2011
- Received: 21 June 2011
- Accepted: 14 December 2011
- Published: 14 December 2011
Abstract
Background
Elevated transient ischemic ST segment episodes in the ambulatory electrocardiographic (AECG) records appear generally in patients with transmural ischemia (e. g. Prinzmetal's angina) while depressed ischemic episodes appear in patients with subendocardial ischemia (e. g. unstable or stable angina). Huge amount of AECG data necessitates automatic methods for analysis. We present an algorithm which determines type of transient ischemic episodes in the leads of records (elevations/depressions) and classifies AECG records according to type of ischemic heart disease (Prinzmetal's angina; coronary artery diseases excluding patients with Prinzmetal's angina; other heart diseases).
Methods
The algorithm was developed using 24-hour AECG records of the Long Term ST Database (LTST DB). The algorithm robustly generates ST segment level function in each AECG lead of the records, and tracks time varying non-ischemic ST segment changes such as slow drifts and axis shifts to construct the ST segment reference function. The ST segment reference function is then subtracted from the ST segment level function to obtain the ST segment deviation function. Using the third statistical moment of the histogram of the ST segment deviation function, the algorithm determines deflections of leads according to type of ischemic episodes present (elevations, depressions), and then classifies records according to type of ischemic heart disease.
Results
Using 74 records of the LTST DB (containing elevated or depressed ischemic episodes, mixed ischemic episodes, or no episodes), the algorithm correctly determined deflections of the majority of the leads of the records and correctly classified majority of the records with Prinzmetal's angina into the Prinzmetal's angina category (7 out of 8); majority of the records with other coronary artery diseases into the coronary artery diseases excluding patients with Prinzmetal's angina category (47 out of 55); and correctly classified one out of 11 records with other heart diseases into the other heart diseases category.
Conclusions
The developed algorithm is suitable for processing long AECG data, efficient, and correctly classified the majority of records of the LTST DB according to type of transient ischemic heart disease.
Keywords
- Ischemic Heart Disease
- Statistical Moment
- Heart Beat
- Manual Classification
- Ischemic Episode
Background
Myocardial ischemia is an adverse outcome of pathologies, which compromise blood flow to the myocardium. It is a state when there is insufficient supply of oxygenated blood or the demand for it is too great. This can cause a part of the heart muscle to become electrically inactive, and can lead to acute myocardial infarction, and in worst case even death. On electrocardiogram (ECG) ischemia is manifested as transient change of ST segment level and morphology (transient ischemic ST segment episodes). In addition to these ischemic episodes, transient non-ischemic ST segment changes also appear [1, 2]. These non-ischemic changes include: changes of ST segment level and morphology due to changes in heart rate (heart-rate related ST segment episodes); sudden changes of ST segment level due to sudden shifts of the electrical axis of the heart (axis shifts) or changes in ventricular conduction (conduction changes); and slow drifts of ST segment level due to diurnal changes or effects of medications. According to shift of ST segment level (positive or negative), transient ischemic ST segment episodes are either elevated or depressed.
Transient ischemic ST segment elevations typically appear in patients with acute transmural ischemia [1] and in patients where acute transmural ischemia without infarction occurs in the settings of Prinzmetal's angina [3, 4]. Furthermore, persistent ST segment elevations indicate higher risk of mortality [5–10], a possible myocardial injury [11–13], and often (but not always) an acute myocardial infarction [14]. In some patients with ST segment elevations, reciprocal ST segment depressions may appear in leads that are separate and distinct from leads manifesting ST segment elevations [1, 4]. These reciprocal changes can appear in leads reflecting contra lateral surface of the heart [1, 4] or are believed to be secondary to coexisting distant ischemia, a manifestation of infarct extension, or may be an electrophysiological phenomenon caused by displacement of the injury current vector away from the non-infarcted myocardium [15]. In contrast to the transient ST segment elevations, the transient ST segment depressions appear in patients with other heart diseases, such as classic stable and unstable angina [3, 4].
Monitoring of a patient over a prolonged time is needed in order to identify or observe sporadic transient events and to asses the extent and severeness of ischemic heart disease. The long-term ambulatory ECG (AECG) records thus obtained have to be checked and analyzed. Huge amount of data dictates use of automated methods for processing and evaluation of such records. As a diagnostic tool for a cardiologist it would be useful, time-saving and helpful to automatically determine "deflections of leads" of AECG records: positive (only elevated transient ischemic ST segment episodes present), negative (only depressed ischemic episodes), mixed (elevated and depressed ischemic episodes), or no deflection (no ischemic episodes); and then to automatically classify records according to "type of ischemic heart disease" into one of three categories: Prinzmetal's angina (PMA), coronary artery diseases excluding patients with Prinzmetal's angina (CAD*), and other heart diseases (OHD). This information could suggest further course of action such as additional investigations or might trigger an early treatment of a patient. The objective of this study was to develop an algorithm to automatically determine deflections of leads of AECG records according to type of transient ischemic ST segment episodes present, and then to automatically classify records according to type of ischemic heart disease into three categories (PMA, CAD*, and OHD).
Methods
For this study we used the AECG records of the Long-Term ST Database (LTST DB) [16], which is intended to develop and to evaluate automated ischemia detectors and to study physiological mechanisms responsible for myocardial ischemia. The LTST DB contains 68 2- and 18 3-lead 24-hour AECG records (altogether 190 AECG leads) sampled at a constant rate of 250 samples per second per channel (ΔT = 4 ms), with amplitude resolution of 200 levels per 1 mV. The records of the LTST DB underwent a considerable preprocessing phase [16], which was essential for human expert annotators to be able to derive reliable reference annotations. Each lead of the records contains reference annotations for transient ischemic and transient non-ischemic heart-rate related ST segment episodes, reference annotations for axis shifts, and reference annotations that define time-varying ST segment reference level (non-ischemic path) along the leads of the records [16]. By subtracting time-varying ST segment reference level, r _{ A } (i, j), where i denotes the lead number and j denotes the heart beat number, from actual ST segment level, s _{ A } (i, j), the ST segment deviation level (or the ST segment deviation function), d _{ A } (i, j), was obtained for each lead of the records. All these functions are stored in the files of the LTST DB. Transient ischemic and transient non-ischemic heart-rate related ST segment episodes were then annotated in the ST segment deviation functions by human expert annotators of the LTST DB. Transient ST segment episodes were annotated according to three annotation criteria. These criteria state that the episode begins when the magnitude of the ST segment deviation function first exceeds 50 μ V. Next, the ST segment deviation function must reach a magnitude of at least V _{ min } μ V throughout the interval of at least T _{ min } s. The episode ends when the ST segment deviation function becomes lower than 50 μ V, provided that it does not exceed 50 μ V in the following 30 s. Values for V _{ min } and T _{ min } differ according to three annotation protocols and are: V _{ min } = 75 μ V and T _{ min } = 30 s for the protocol A; V _{ min } = 100 μ V and T _{ min } = 30 s for the protocol B; and V _{ min } = 100 μ V and T _{ min } = 60 s for the protocol C.
Manual classification of the records of the LTST DB
LTST DB (74 records) | |||||
---|---|---|---|---|---|
Type of heart disease ↓ | Diagnoses | ||||
Prinzmetal's angina | Unstable angina | Angina | CAD | Other | |
PMA | 6 | 1 | 1 | 0 | 0 |
CAD* | 1 | 5 | 4 | 39 | 3 |
OHD | 1 | 1 | 0 | 4 | 8 |
The developed algorithm to classify AECG records into the categories of type of ischemic heart disease is composed from three modules: A. preprocessing; B. tracking of slow drifts, detection of axis shifts and correcting the ST segment reference level; and C. determining the deflection of leads and classifying the records according to type of ischemic heart disease.
A. Preprocessing
The input to the developed algorithm were raw AECG data of the records and the ARISTOTLE's [17] fiducial points of normal and non-noisy heart beats which passed the preprocessing phase of the LTST DB [16]. These data are freely available to the users of the LTST DB. To further avoid the effects of noise, the average heart beats were constructed. Each normal heart beat in the raw AECG signal was replaced with the average heart beat. For the construction of each average heart beat normal non-noisy heart beats in the 16 s neighborhood of the current heart beat were used. Heart beats were aligned according to their fiducial points, FP(i, j), where i denotes lead number, and j denotes heart beat number. The FP(i, j) is located in QRS complex region of the j-th heart beat in the 'center of mass' of deflections [16].
To construct the ST segment level function, the algorithm searches for the positions of the isoelectric level and J point in each average heart beat. To determine the position of the isoelectric level of the j-th heart beat, I(j), the algorithm searches from the FP(i, j) backwards to point FP(i, j) - 108 ms in each lead for the "flattest" 20 ms segment of the signal and then determines one final position, I(j), [18, 19]. For the position of the J point of the j-th heart beat, J(j), the algorithm searches forward from the FP(i, j) to the point FP(i, j) + 68 ms in each lead for a part of a waveform which "starts to flatten". One final position, J(j), is then determined as that furthest from the FP(i, j) [19, 20].
B. Tracking of slow drifts, detection of axis shifts and correcting the ST segment reference level
In order to accurately determine deflection of leads, all non-ischemic events have to be excluded from further analysis. For this, the algorithm tracks the time-varying non-ischemic path in each ST segment level function along the record to create the ST segment reference function. The ST segment reference function is then subtracted from the ST segment level function to obtain the ST segment deviation function, which is used to determine deflection of leads. Construction of the ST segment reference function is made in several steps.
In the first step, the algorithm tracks slowly varying ST segment reference level trend by applying two moving average filters of 6 h 40 min (h _{ g } ) and 5 min (h _{ l } ) in length to the ST segment level function, s(i, k), to obtain the time-varying global, r _{ g } (i, k), and local, r _{ l } (i, k), ST segment reference level trends, respectively. The lengths of the impulse responses of the filters were selected such that the output of h _{ g } models slow changes of the ST segment level function (e. g. slow drifts), while output of h _{ l } models faster changes of the ST segment level function (e. g. transient ST segment episodes) [21]. Using these two ST segment reference level trends, the algorithm obtains the estimation of the ST segment reference function, r _{1}(i, k), which tracks slow drifts but skips faster events and episodes. The r _{1}(i, k) is taken as r _{ g } (i, k) if the absolute difference between the r _{ g } (i, k) and r _{ l } (i, k) is more than 50 μ V; otherwise it is taken as r _{ l } (i, k).
Ideally, we would get the ST segment deviation function where only transient ST segment episodes are present. The obtained ST segment reference function of the example from Figure 1 (Figure 1e) tracks the non-ischemic changes quite well, so the ST segment deviation function derived by the algorithm (Figure 1f) is quite similar to that constructed by the human expert annotators of the LTST DB (Figure 1c).
C. Determining the deflection of leads and classifying the records according to type of ischemic heart disease
where K _{ C } is the threshold for lead classification and differs according to the statistical moment used. The threshold K _{ c } is the same for all leads of all records of the database, given statistical moment used for lead classification (either the first, or the second, or the third). The example in Figure 2 demonstrates determining deflection of the first lead of the record s20371 using the third statistical moment.
To optimize the algorithm, we investigated the first, second, and third statistical moment, for various values of threshold K _{ C } . The K _{ C } determines whether deflection of a lead is decided as positive, or negative, or as mixed or no episodes. Using higher K _{ C } more leads have deflections of leads determined as mixed or no episodes, while using lower K _{ C } more leads have deflections of leads determined as positive or negative.
As the main optimization constraint we took the minimum number of leads containing only elevated ischemic episodes being falsely determined as having negative, and the minimum number of leads containing only depressed ischemic episodes being falsely determined as having positive deflections of leads. We also tried to maximize the number of leads containing both types of ischemic episodes and leads without any ST segment episodes determined as having mixed or no episodes deflections of leads. We tested different values of the threshold K _{ C } for the first, second, and third statistical moment. The optimal values obtained for the threshold K _{ C } were: 2 × 10^{3}(μ V) for the first, 75 × 10^{3}(μ V)^{2} for the second, and 3.75 × 10^{6}(μ V)^{3} for the third statistical moment.
Finally, the algorithm automatically classifies each record p into one of the categories of type of ischemic heart disease, D _{ p } ∈ {PMA, CAD*, OHD}, using the set of rules (1).
Results
Results of determining deflections of leads
Reference level correction | No reference level correction | |||||
---|---|---|---|---|---|---|
First moment | K _{ C } = 2 × 10 ^{ 3 } ( μ V) | K _{ C } = 3.5 × 10 ^{ 3 } ( μ V) | ||||
positive | negative | mixed, no episodes | positive | negative | mixed, no episodes | |
Elevations | 9 (90%) | 0 (0%) | 1 (10%) | 8 (80%) | 2 (20%) | 0 (0%) |
Depressions | 3 (3%) | 89 (96%) | 1 (1%) | 24 (26%) | 67 (72%) | 2 (2%) |
Mixed | 3 (60%) | 2 (40%) | 0 (0%) | 2 (40%) | 3 (60%) | 0 (0%) |
No episodes | 11 (26%) | 24 (56%) | 8 (19%) | 11 (26%) | 27 (63%) | 5 (12%) |
Second moment | K _{ C } = 75 × 10 ^{ 3 } ( μ V) ^{ 2 } | K _{ C } = 150 × 10 ^{ 3 } ( μ V) ^{ 2 } | ||||
positive | negative | mixed, no episodes | positive | negative | mixed, no episodes | |
Elevations | 10 (100%) | 0 (0%) | 0 (0%) | 9 (90%) | 1 (10%) | 0 (0%) |
Depressions | 0 (0%) | 90 (97%) | 3 (3%) | 17 (18%) | 76 (82%) | 0 (0%) |
Mixed | 3 (60%) | 2 (40%) | 0 (0%) | 2 (40%) | 3 (60%) | 0 (0%) |
No episodes | 10 (23%) | 20 (47%) | 13 (30%) | 12 (28%) | 24 (56%) | 7 (16%) |
Third moment | K _{ C } = 3.75 × 10 ^{ 6 } ( μ V) ^{ 3 } | K _{ C } = 4.75 × 10 ^{ 6 } ( μ V) ^{ 3 } | ||||
positive | negative | mixed, no episodes | Positive | negative | mixed, no episodes | |
Elevations | 10 (100%) | 0 (0%) | 0 (0%) | 9 (90%) | 1 (10%) | 0 (0%) |
Depressions | 0 (0%) | 92 (99%) | 1 (1%) | 12 (13%) | 81 (87%) | 0 (0%) |
Mixed | 3 (60%) | 2 (40%) | 0 (0%) | 2 (40%) | 3 (60%) | 0 (0%) |
No episodes | 9 (21%) | 17 (40%) | 17 (40%) | 13 (30%) | 24 (56%) | 6 (14%) |
The lower left part of Table 2 shows the results of determining deflections of leads using the third statistical moment. The algorithm correctly determined deflections of 100% of leads with elevations as positive, and of 99% of leads with depression as negative. For the leads with mixed types of episodes the algorithm determined deflections of 60% of leads as positive and of 40% as negative. The algorithm also correctly determined deflections of 40% of leads without transient ST segment episodes as mixed or no episodes. The right part of Table 2 shows results of determining deflections of leads using different statistical moments when no reference level correction was performed. Values of the threshold K _{ C } were optimized for best performance. Using the first statistical moment (upper right, refer also to left part of the Table 2), the algorithm correctly determined deflections of 80% of leads with elevations as positive and of 72% of leads with depressions as negative. The algorithm also correctly determined deflections of 12% of leads without transient ST segment episodes as mixed or no episodes. Using the second statistical moment (middle right), the algorithm correctly determined deflections of 90% of leads with elevations as positive and of 82% of leads with depressions as negative. The algorithm also correctly determined deflections of 16% of leads without transient ST segment episodes as mixed or no episodes. Finally, using the third statistical moment (lower right), the algorithm correctly determined deflections of 90% of leads with elevations as positive and of 87% of leads with depressions as negative. The algorithm also correctly determined deflections of 14% of leads without transient ST segment episodes as mixed or no episodes.
Manual and automatic classification (with differences) of the records of the LTST DB
Reference: LTST DB (74 records) | ||||||
---|---|---|---|---|---|---|
Type of heart disease ↓ | Diagnoses | |||||
Prinzmetal's angina | Unstable angina | Angina | CAD | Other | ||
PMA | 6 | 1 | 1 | 0 | 0 | |
CAD* | 1 | 5 | 4 | 39 | 3 | |
OHD | 1 | 1 | 0 | 4 | 8 | |
Σ8 | Σ55 | Σ11 | ||||
Algorithm: LTST DB (74 records) | ||||||
Type of heart disease ↓ | Diagnoses | |||||
Prinzmetal's angina | Unstable angina | Angina | CAD | Other | ||
PMA | [7] Σ7 | [2 | 2 | 2] | Σ6 | 6 |
CAD* | 1 | [5 | 3 | 39] | Σ47 | 4 |
OHD | 0 | [0 | 0 | 2] | Σ2 | [1] Σ2 |
Σ8 | Σ 55 | Σ11 | ||||
Algorithm - Reference | ||||||
Type of heart disease ↓ | Diagnoses | |||||
Prinzmetal's angina | Unstable angina | Angina | CAD | Other | ||
PMA | 1 | 1 | 1 | 2 | 6 | |
CAD* | 0 | 0 | -1 | 0 | 1 | |
OHD | -1 | -1 | 0 | -2 | -7 |
The algorithm (middle) correctly classified 7 out of 8 patients with Prinzmetal's angina into PMA category, and one was misclassified into CAD* category. The algorithm classified 47 out of 55 patients with unstable angina, angina, or coronary artery disease into CAD* category, while 6 were misclassified into PMA and two into OHD category. One patient out of 11 with other heart disease was classified into OHD category. The difference between automatic and manual classification (bottom) shows, that the automatic classification gives pretty much similar results as the manual classification. Using the set of rules (1) the algorithm managed to correctly recognize majority of records belonging to patients with Prinzmetal's angina and majority of records belonging to patients with other coronary artery diseases.
Discussion and Conclusions
Our algorithm performed exceptionally well in determining deflections of leads and in classifying patients according to type of ischemic heart disease, but there are still some limitations, which concern leads containing both types of transient ischemic ST segment episodes and leads without transient ST segment episodes. In cases where leads contain larger number of one type of ischemic episodes, the insertion of such leads into mixed group seems to be inadequate. A division of these records into more groups, or some other method for determining deflections of leads, might be considered, and the rule (7) for determining deflections of leads would need to be improved. The problem concerning leads without transient ST segment episodes is the inability of the algorithm to accurately detect all axis shifts. To rectify this, the developed method for detecting axis shifts would need to be improved. Other techniques for detecting axis shifts due to body position changes were investigated in the past [22–24]. Pitfalls with these techniques lie in the fact that they were developed using artificially triggered axis shifts, so prior validation of these techniques using real clinical data should be performed.
The algorithm shows high sensitivity of determining deflection of leads (100% for leads containing elevations only and 99% for leads containing depressions only) with some false positives. The proposed algorithm is efficient and could be a valuable aid in every day clinical practice. The algorithm is, despite some limitations, appropriate for processing large amount of AECG data and for quick assessment of type of ischemic heart disease. The study showed that the reference level correction (tracking of slow drifts, detection of axis shifts, and correcting the ST segment reference level) is an essential part of the algorithm and enables good classification of patients according to type of ischemic heart disease. Without the module for reference level correction, the deflections of a substantial part of leads were incorrectly determined. Early and accurate assessment of the deflection of leads itself is already valuable for a clinician, since this information suggests the cause and type of ischemia.
Declarations
Acknowledgements
No outside funding was received for this study.
Authors’ Affiliations
References
- Braunwald E, Zipes DP, Libby P: Heart disease: a textbook of cardiovascular medicine. 9th edition. Philadelphia, PA, USA: W. B. Saunders Company; 2011.Google Scholar
- Hampton JR: The ECG in practice. 4th edition. Oxford, UK: Churchill Livingstone; 2003.Google Scholar
- Goldberger A: Electrocardiograms in Ischemic Heart Disease. West J Med 1981, 135: 104–109.Google Scholar
- Goldberger AL: Clinical Electrocardiography: A Simplified Approach. 7th edition. Philadelphia, PA, USA: Mosby Elsevier; 2006.Google Scholar
- Quintana M, Lindvall K, Brolund F: Assessment and Significance of ST-Segment Changes Detected by Ambulatory Electrocardiography After Acute Myocardial Infarction. Am J Cardiol 1995, 76: 6–13. 10.1016/S0002-9149(99)80792-9View ArticleGoogle Scholar
- Kleiman NS, Granger CB, White HD, Armstrong P, Ardissino D, Van de Werf F, Zoldeyhi P, Thompson TD, Califf RM, Topol EJ: Death and nonfatal reinfarction within the first 24 hours after presentation with an acute coronary syndrome: Experience from GUSTO-IIb. Am Heart J 1999, 137: 12–23. 10.1016/S0002-8703(99)70455-2View ArticleGoogle Scholar
- Udagawa H, Yoshino H, Kachi E, Taniuchi M, Yotsukura M, Ishikawa K: ST-Segment Elevation in Leads I and aVL Predicts Short-Term Prognosis in Acute Anterior Wall Myocardial Infarction. Am J Cardiol 2000, 85: 101–104. 10.1016/S0002-9149(99)00615-3View ArticleGoogle Scholar
- Maas AC, Wyatt CM, Green CL, Wagner GS, Trollinger KM, Pope JE, Langer A, Armstrong PW, Califf RM, Simoons ML, Krucoff MW: Combining baseline clinical descriptors and real-time response to therapy: The incremental prognostic value of continuous ST-segment monitoring in acute myocardial infarction. Am Heart J 2004, 147: 698–704. 10.1016/j.ahj.2003.08.014View ArticleGoogle Scholar
- Pitta SR, Grzybowski M, Welch RD, Frederick PD, Wahl R, Zaleski RJ: ST-segment depression on the initial electrocardiogram in acute myocardial infarction-prognostic significance and its effect on short-term mortality: A report from the National Registry of Myocardial Infarction (NRMI-2, 3, 4). Am J Cardiol 2005, 95: 843–848. 10.1016/j.amjcard.2004.12.010View ArticleGoogle Scholar
- Wong CK, French JK, Aylward PE, Stewart RAH, Gao W, Armstrong PW, Van De Werf FJ, Simes RJ, Raffel OC, Granger CB, Califf RM, White HD: Patients With Prolonged Ischemic Chest Pain and Presumed-New Left Bundle Branch Block Have Heterogeneous Outcomes Depending on the Presence of ST-Segment Changes. J Am Coll Cardiol 2005, 46: 29–38. 10.1016/j.jacc.2005.02.084View ArticleGoogle Scholar
- Birnbaum Y, Hasdai D, Sclarovsky S, Herz I, Strasberg B, Rechavia E: Acute myocardial infarction entailing ST-segment elevation in lead aVL: Electrocardiographic differentiation among occlusion of the left anterior descending, first diagonal, and first obtuse marginal coronary arteries. Am Heart J 1996, 131: 38–42. 10.1016/S0002-8703(96)90048-4View ArticleGoogle Scholar
- Figueras J, Bermejo B: Additional elevation of the ST segment: a possible early electrocardiographic marker of experimental myocardial reperfusion. Cardiovasc Res 1996, 32: 1141–1147. 10.1016/S0008-6363(96)00161-7View ArticleGoogle Scholar
- Kleber AG: ST segment elevation in electrocardiogram: a sign of myocardial ischemia. Cardiovasc Res 2003, 45: 111–118.View ArticleGoogle Scholar
- Wang K, Asinger RW, Marriott HJL: ST Segment Elevation in Conditions Other Than Acute Myocardial Infarction. N Engl J Med 2003, 349: 2128–2135. 10.1056/NEJMra022580View ArticleGoogle Scholar
- Pollehn T, Brady WJ, Perron AD, Morris F: The electrocardiographic differential diagnosis of ST segment depression. Emerg Med J 2002, 19: 129–135. 10.1136/emj.19.2.129View ArticleGoogle Scholar
- Jager F, Taddei A, Moody GB, Emdin M, Antolič G, Dorn R, Smrdel A, Marchesi C, Mark RG: Long-term ST database: a reference for the development and evaluation of automated ischaemia detectors and for the study of the dynamics of myocardial ischaemia. Med Biol Eng Comput 2003, 41: 172–182. 10.1007/BF02344885View ArticleGoogle Scholar
- Moody GB, Mark RG: Development and evaluation of a 2-lead ECG analysis program. IEEE Computers in Cardiology 1982 1982, 9: 39–44.Google Scholar
- Jager F, Mark RG, Moody GB, Divjak S: Detection of transient ST-segment episodes during ambulatory ECG-monitoring. Comput Biomed Res 1998, 31: 305–322. 10.1006/cbmr.1998.1483View ArticleGoogle Scholar
- Smrdel A, Jager F: An algorithm to estimate the ST Segment level in 24-hour ambulatory ECG records. IEEE Computers in Cardiology 2008 2008, 35: 701–704.View ArticleGoogle Scholar
- Daskalov IK, Dotsinsky IA, Christov II: Development in ECG acquisition, preprocessing, parameter measurement, and recording. IEEE Eng Med Biol Mag 1998, 17: 50–58. 10.1109/51.664031View ArticleGoogle Scholar
- Smrdel A, Jager F: Automated detection of transient ST-segment episodes in 24 h electrocardiograms. Med Biol Eng Comput 2004, 42: 303–311. 10.1007/BF02344704View ArticleGoogle Scholar
- Åström M, Garcia J, Laguna P, Pahlm O, Sörnmo L: Detection of body position changes using the surface electrocardiogram. Med Biol Eng Comput 2003, 41: 164–171. 10.1007/BF02344884View ArticleGoogle Scholar
- Garcia J, Åström M, Mendive J, Laguna P, Sörnmo L: ECG-Based Detection of Body Position Changes in Ischemia Monitoring. IEEE Trans Biomed Eng 2003, 50: 677–685. 10.1109/TBME.2003.812208View ArticleGoogle Scholar
- Shinar Z, Bahrav A, Akselrod S: Detection of different recumbent body positions from the electrocardiogram. Med Biol Eng Comput 2003, 41: 206–210. 10.1007/BF02344890View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.