Gaël de Lannoy

Supervised ECG Delineation Using the Wavelet Transform and Hidden Markov Models

Clinical monitoring and pharmaceutical phaseone studies require feature extraction from the ECG signal in order to evaluate the state of a patient’s heart. Automatic annotation of the characteristic ECG waveforms (or so-called delineation) is therefore of great interest. Hidden Markov Models (HMM) coupled to wavelet transforms (WT) of the ECG signal offer significant improvements over standard heuristic delineation methods. Nevertheless, the choice of the WT parameters remains empirical rather than data-driven. In these conditions, suboptimal parameters for the WT may degrade the results very much. In this paper, an algorithm for the optimal selection of the WT parameter values is introduced. The model complexity is strongly reduced and the algorithm can adapt itself to the specificities of each ECG signal while avoiding redundancy, noise and useless information. Evaluation on recordings from the public MITQT database leads to results higher than with state of the art methods.

Weighted SVMs and Feature Relevance Assessment in Supervised Heart Beat Classification.

The diagnosis of cardiac dysfunctions requires the analysis of long-term ECG signal recordings, often containing hundreds to thousands of heart beats. In this work, automatic inter-patient classification of heart beats following AAMI guidelines is investigated. The prior of the normal class is by far larger than the other classes, and the classifier obtained by a standard SVM training is likely to act as the trivial acceptor. To avoid this inconvenience, a SVM classifier optimizing a convex approximation of the balanced classification rate rather than the standard accuracy is used. First, the assessment of feature sets previously proposed in the litterature is investigated. Second, the performances of this SVM model is compared to those of previously reported inter-patient classification models. The results show that the choice of the features is of major importance, and that some previously reported feature sets do not serve the classification performances. Also, the weighted SVM model with the best feature set selection achieves results better than previously reported inter-patient models with features extracted only from the R spike annotations.

Label noise-tolerant hidden Markov models for segmentation: application to ECGs

The performance of traditional classification models can adversely be impacted by the presence of label noise in training observations. The pioneer work of Lawrence and Scholkopf tackled this issue in datasets with independent observations by incorporating a statistical noise model within the inference algorithm. In this paper, the specific case of label noise in non-independent observations is rather considered. For this purpose, a label noise-tolerant expectation-maximisation algorithm is proposed in the frame of hidden Markov models. Experiments are carried on both healthy and pathological electrocardiogram signals with distinct types of additional artificial label noise. Results show that the proposed label noise-tolerant inference algorithm can improve the segmentation performances in the presence of label noise.

A Supervised Wavelet Transform Algorithm for R Spike Detection in Noisy ECGs

The wavelet transform is a widely used pre-filtering step for subsequent R spike detection by thresholding of the coefficients. The time-frequency decomposition is indeed a powerful tool to analyze non-stationary signals. Still, current methods use consecutive wavelet scales in an a priori restricted range and may therefore lack adaptativity. This paper introduces a supervised learning algorithm which learns the optimal scales for each dataset using the annotations provided by physicians on a small training set. For each record, this method allows a specific set of non consecutive scales to be selected, based on the record’s characteristics. The selected scales are then used for the decomposition of the original long-term ECG signal recording and a hard thresholding rule is applied on the derivative of the wavelet coefficients to label the R spikes. This algorithm has been tested on the MIT-BIH arrhythmia database and obtains an average sensitivity rate of 99.7% and average positive predictivity rate of 99.7%.

Emission Modelling for Supervised ECG Segmentation using Finite Differences

The segmentation of ECG signals into P waves, QRS complexes, T waves and baselines is an important practical problem for physicians diagnosing cardiac diseases. The duration of the signal and the number of beats to segment are often too large for a manual annotation, so that automatic segmentation is a challenging and useful tool. State-of-the-art algorithms use hidden Markov models with wavelet transform encoding and represent the ECG in multidimensional spaces using Gaussian mixtures models. The main problem of this approach is its computational cost due to the number of free parameters, the choice of the wavelet transform parameters and the high failure rate of the EM algorithm. In this work, we propose an alternative emission encoding for hidden Markov models using both the ECG signal and its derivative in order to better model the dynamics of the signal in a lower dimensional space. We show that this method achieves similar performances with much less model parameters and is less subject to failures.