Karhunen-Loève transform as a tool to analyze the ST-segment. Comparison with QT interval

A statistically based acute ischemia detection algorithm suitable for an implantable device

This study investigates the performance of a new statistically driven acute ischemia detection algorithm that can process data from two bipolar cutaneous or subcutaneous leads. During a start-up phase, the algorithm processes electrocardiogram signals to determine a normal range of ST-segment deviation as a function of heart rate. The algorithm then generates upper and lower ST-deviation thresholds based on the dispersion of the baseline ST-deviation data. After the start-up phase, persistent ST-deviation that is beyond either the upper or lower thresholds results in detection of acute ischemia. To test the algorithm, we performed long-term (10 day) Holter monitoring in a control group of 14 subjects. We also performed Holter monitoring during balloon angioplasty, and for 2 days after surgery, in 30 subjects who underwent elective percutaneous coronary interventions ("PCI"). We determined the percentage of balloon inflations the algorithm detected without producing false positive detections within the control group 10-day daily life data. The algorithm detected 17/17 LAD occlusions, 7/8 LCX occlusions, and 8/9 RCA occlusions. Our results suggest that automatically generated, subject-specific, heart-rate dependent ST-deviation thresholds can detect PCI induced myocardial ischemia without resulting in false positive detections in a small control group.

Measurement of high resolution ECG QT interval during controlled euglycaemia and hypoglycaemia

During hypoglycaemia, typically there is a change in the surface ECG characterized by a flattened and prolonged T wave, often accompanied by a fused U wave. The QT interval is a useful parameter for quantifying the ECG morphology. However, reliable measurement of QT is not straightforward, particularly for hypoglycaemic ECG morphology. The objective of this study was to compare the ability of two methods of QT measurement to distinguish between ECGs recorded during euglycaemia and hypoglycaemia. The first method involves manually setting the intersection of the isoelectric line and the T wave or, where this is not possible, the nadir between the T and U wave. The second method is semi-automatic and fits a tangent to the point of maximum gradient on the downward slope of the T wave. Two independent observers used both methods to measure the QT for high resolution ECG data recorded during a study of 17 non-diabetic subjects undergoing controlled euglycaemia and hypoglycaemia. Using the mean results of the two observers, the mean +/- SD increase in heart rate corrected QT, QTc, for ECGs recorded during euglycaemia and hypoglycaemia was 32 +/- 25 ms for the non-tangent method and 60 +/- 24 ms for the tangent method. Therefore, the tangent method provides greater distinction between ECGs recorded during euglycaemia and hypoglycaemia than the non-tangent method. A potential clinical application could be the non-invasive detection of impending hypoglycaemia at night, which would be of significant benefit to adults and young children with diabetes.

Temporal evolution of traditional versus transformed ECG-based indexes in patients with induced myocardial ischemia

The time course of changes in the electrocardiogram as a result of myocardial ischemia induced during prolonged coronary angioplasty has been studied. We have analyzed the electrocardiogram evolution during the occlusion in terms of the Ischemic Changes Sensor, which is a parameter that describes the capacity of different indexes to detect induced changes. Traditional indexes at specific time locations (ST level, T wave amplitude and position, and durations of QT interval and QRS complex) and global indexes (based on the Karhunen-Loève transform as applied to the QRS complex, ST-T complex, ST segment and T wave) have been considered. The global indexes better detected ischemic changes than the traditional indexes. The most sensitive were the index for the ST-T complex (89%) in the Karhunen-Loève transform-derived group and for the ST level (61%) in the traditional group. Changes in the ventricular repolarization period usually appeared earlier (77% of patients) than changes in the depolarization period (23% of patients). A similar percentage of patients exhibited the earliest ischemic changes in the T wave (41%) and in the ST segment (36%). The evolution of the Ischemic Changes Sensor parameters showed that the majority (60%) of the total changes occurred during the first minute of occlusion. The results suggest that the use of global electrocardiogram indexes better reflect ischemic changes than do traditional indexes, such as the ST segment deviation.

Identification of the occluded artery in patients with myocardial ischemia induced by prolonged percutaneous transluminal coronary angioplasty using traditional vs transformed ECG-based indexes

We have studied the spatial properties of ischemic changes as induced by prolonged angioplasty and how the changes are related to different ECG indexes. Indexes based on measurements at specific points in time (ST level at J + 60 ms point, maximal T wave amplitude and position, QT interval, and QRS duration) and global indexes (based on the Karhunen-Loève transform and applied to the QRS complex, ST-T complex, ST segment, and T wave), considering both repolarization and depolarization information, were analyzed. The changes during the occlusion period of the different indexes were used as variables in a multivariate discriminant analysis to determine which indexes showed the best discrimination of the three major occlusion sites (corresponding to LAD, RCA, and LCX coronary arteries). Occlusions in LCX artery were the most difficult to classify. With three local indexes (ST60 level measured in lead V3, T wave amplitude in I, and ST60 in III) it was possible to correctly classify 76% of patients by the occlusion site, and with three KLT-derived indexes (first-order KLT index for ST-T complex in I and for QRS in leads V3 and I) 83% of correct classification was obtained. Using six indexes for local and KLT-derived indexes the correct classification was increased to 85 and 90% of patients, respectively. The use of different ECG indexes (from different intervals) on quasiorthogonal leads permitted the identification of the occluded artery in patients undergoing PTCA and may be extended to more general use.

Spatial, temporal and wavefront direction characteristics of 12-lead T-wave morphology

Three new approaches for the analysis of ventricular repolarisation in 12-lead electrocardiograms (ECGs) are presented: the spatial and temporal variations in T-wave morphology and the wavefront direction difference between the ventricular depolarisation and repolarisation waves. The spatial variation characterises the morphology differences between standard leads. The temporal variation measures the change in interlead relationships. A minimum dimensional space, constructed by ECG singular value decomposition, is used. All descriptors are measured using the ECG vector in the constructed space and the singular vectors that define this space. None of the descriptors requires time domain measurements (e.g. the precise detection of the T-wave offset), and so the inaccuracies associated with conventional QT interval related parameters are avoided. The new descriptors are compared with the conventional measurements provided by a commercial system for an automatic evaluation of QT interval and QT dispersion in digitally recorded 12-lead ECGs. The basic comparison uses a set of 1100 normal ECGs. The short-term intrasubject reproducibility of the new descriptors is compared with that of the conventional measurements in a set of 760 ECGs recorded in 76 normal subjects and a set of 630 ECGs recorded in 63 patients with hypertrophic cardiomyopathy (ten serial recordings in each subject of both these sets). The discriminative power of the new and conventional parameters to distinguish normal and abnormal repolarisation patterns is compared using the same set. The results show that the new parameters do not correlate with the conventional QT interval-related descriptors (i.e. they assess different ECG qualities), are generally more reproducible than the conventional parameters, and lead to a more significant separation between normal and abnormal ECGs, both univariately and in multivariate regression models.

Similarity measurement method for the classification of architecturally differentiated images

A similarity measurement method for the classification of architecturally differentiated image sections is described. The strength of the method is demonstrated by performing the complex task of assigning severity grading (Gleason grading) to histological slides of prostate cancer. As shown, all that is required to employ the method is a small set of preclassified images. The images can be real world images acquired by means of a camera, computer tomography, etc., or schematic drawings representing samples of different classes. The schematic option allows a quick test of the method for a particular classification problem.

Comparative study of local and Karhunen-Loève-based ST-T indexes in recordings from human subjects with induced myocardial ischemia

In this work we studied ST-T complex changes in the ECG as result of induced ischemia. The principal aim was to determine whether global changes in the ST-T complex were more sensitive markers of ischemic alterations than those based on measurements of changes at specific locations on ST segment or T wave. High-resolution ECGs from patients undergoing percutaneous transluminal coronary angioplasty in one of the major coronary arteries were analyzed to give a description of the period from the end of active depolarization (QRS complex) to the end of active repolarization (T wave). During artery occlusion traditional local measurements of the ST-T complex were compared to global measurements based on the Karhunen-Loève transform. An ischemic change sensor parameter was estimated for each of the studied indexes showing that global measurements detected changes better in the repolarization period in a larger number of leads and with higher sensitivity (more than 85%) than was done using local measurements (sensitivity of 64% with ST level, 33% with T-wave maximum position, and 37% with T-wave maximum amplitude). Using these global indexes it was found that most cases of ST-segment changes were accompanied by T-wave changes (72% of patients). With the use of traditional indexes 23% of patients showed no changes in the repolarization period, whereas with global indexes this percentage decreased to 8%. Thus a global representation of the entire ST-T complex appears to be more suitable than local measurements when studying the initial stages of myocardial ischemia.

Model-based estimation of cardiovascular repolarization features: ischaemia detection and PTCA monitoring

The ST-T segment of the surface ECG reflects cardiac repolarization, and is quite sensitive to a number of pathological conditions, particularly ischaemia. ST-T changes generally affect the entire waveshape, and are inadequately characterized by single features such as depression of the ST segment at one particular point. Metrics which represent overall waveshape should provide more sensitive indicators of ST-T wave abnormalities, particularly when they are subtle, intermittent or periodic. This study discusses a Karhunen-Loève transform (KLT) technique for the analysis of the ST-T waveform. The KL technique was used to analyse the ST-T complexes in the ESC ST-T database. KL coefficients were plotted as a function of time, and were effective in detection of transient ischaemic episodes. Twenty per cent of the records showed bursts of periodic ischaemia suggesting local vascular instability. A comparison between kl and ST depression series has shown the KL technique as more appropriate to the study of ST-T complex variations. Using the kl series, an ischaemia detector has been developed based on a resampled, filtered, and differentiated KL series. This technique demonstrates a sensitivity of 65% and a specificity of 54%. These low values can be due to shifts of the electrical axis which are detected as ischaemic changes, real ischaemic episodes that were not annotated with the protocol used at the European ST-T database, or erroneous detections. An increase in sensitivity can be obtained at the expense of a decrease in the positive predictive value and thus becomes a useful technique for previous scanning of the ECG record and subsequent review by the expert. The technique has also been used to monitor patients during a PTCA process, demonstrating that this technique allows us to monitor PTCA-induced ischaemia. A detailed analysis has shown that in some cases a repetitive oscillatory behaviour appears, lasting for a period of around 20 s, and highly related to the oscillatory behaviour of the HR. In other cases, transient changes in KL series with salves behaviour associated with the injection of contrast are shown on the ST-T waveform. We conclude that the KL-based analysis of the ST-T segment is a robust and sensitive technique, with considerable advantages over single feature measures in characterizing the subtle waveform changes which may be of importance in clinical risk detection.