Daniel Schang

Early prediction of unexplained syncope by support vector machines

The goal of the present study was to develop and evaluate a new method for the prediction of unexplained syncope occurrences. Diagnosis of syncope is currently based on the reproduction of symptoms in combination with hypotension and bradycardia induced by a 45 min 60-70 degrees head-upright tilt test (HUTT). The main drawback of this widely used test concerns its duration that reaches 55 min if the patient does not faint. Our method is a first step in the avoidance of the HUTT. An electrocardiogram and a transthoracic impedance waveform were recorded for 10 min of supine rest of a HUTT in 128 patients with a history of unexplained recurrent syncope. Seven indices were computed on the transthoracic impedance and its first derivative. The prediction quality of every subset of these variables, mixed with age and sex, has been tested by a support vector machine in a retrospective group of 64 patients (100% of sensitivity and 100% of specificity was reached). The best subset obtained has been evaluated prospectively in a group of 64 patients (94% of sensitivity and 79% of specificity was reached). These results compare very favorably with published results for other unexplained syncope detectors.

Late Hemodynamic Changes During a Negative Passive Head-up Tilt Predict the Symptomatic Outcome to a Nitroglycerin Sensitized Tilt

Background: Sublingual nitroglycerin is advocated to sensitize the passive 70° head‐upright tilt test (HUTT) of patients with unexplained syncope. We hypothesized that a detailed analysis of hemodynamic responses recorded during a negative HUTT could predict the outcome to a subsequent nitroglycerin sensitized HUTT (NTG‐HUTT).

Prediction of blood transfusion donation

The goal of the present study was to develop and evaluate machine learning algorithms for the prediction of blood transfusion donation. The machine learning algorithms studied included multilayer perceptrons (MLPs) and support vector machines (SVMs). The methods were evaluated retrospectively in a group of 600 patients and validated prospectively in a group of 148 patients. We reach a sensitivity of 65.8% and a specificity of 78.2% in the prospective group. This discrimination is very interesting because it could allow to propose to the patients, classified as non-donators, to give their blood in the future. Furthermore, the blood transfusion donation UCI corpus used, has been processed in a different manner than the initial marketing one. Therefore, this recent corpus could give a new training set for testing and improving machine learning methods in the future.

Evaluation of automatic feature detection algorithms in EEG

In this paper, we present a new method to compare and improve algorithms for feature detection in neonatal EEG. The method is based on the algorithm׳s ability to compute accurate statistics to predict the results of EEG visual analysis. This method is implemented inside a Java software called EEGDiag, as part of an e-health Web portal dedicated to neonatal EEG. EEGDiag encapsulates a component-based implementation of the detection algorithms called analyzers. Each analyzer is defined by a list of modules executed sequentially. As the libraries of modules are intended to be enriched by its users, we developed a process to evaluate the performance of new modules and analyzers using a database of expertized and categorized EEGs. The evaluation is based on the Davies-Bouldin index (DBI) which measures the quality of cluster separation, so that it will ease the building of classifiers on risk categories. For the first application we tested this method on the detection of interburst intervals (IBI) using a database of 394 EEG acquired on premature newborns. We have defined a class of IBI detectors based on a threshold of the standard deviation on contiguous short time windows, inspired by previous work. Then we determine which detector and what threshold values are the best regarding DBI, as well as the robustness of this choice. This method allows us to make counter-intuitive choices, such as removing the 50 Hz filter (power supply) to save time.

Creep test material rupture prediction by neural networks

This work focuses on acoustic emission analysis of mechanisms damage in fiber composite materials, subjected to heavy loads during a creep test. The goal of the present study was to develop and evaluate machine learning algorithms for the prediction of material rupture with creep test by traction method. This study aimed to predict if a tensile specimen will break in 30 seconds or not. Multilayer Perceptrons were trained retrospectively in a group of 80 samples moments and tested prospectively in a group of 16 tensile specimens. During the 5-cross validations we reached a sensitivity of 88% and a specificity of 88% in the prospective group. The mean area under the ROC (Receiver Operating Curves) was equal to 0.86. Those results are very interesting because they are a first important step in the lifetime prediction of material rupture before significant damages can occur.

Dimension reduction methods for the early syncope prediction by artificial neural networks

The aim of this study is to develop a method to predict unexplained syncope or presyncope occurrences. Diagnosis of syncope is currently based on the reproduction of symptoms in combination with hypotension and bradycardia induced by a 45-min of 60-80 head-upright tilt test (HUTT). The main drawback of this test concerns its duration which can reach 55 minutes if the patient does not faint. Thus we propose a new predicting tool which is only based on the 10 first minutes of the supine position of the HUTT. First, we describe how to acquire the variables of the disease and to achieve the dimension reduction methods. Then, at the end of the variables processing a neuronal method evaluates the prediction quality for a retrospective and a prospective group of patients. The best model compares very favorably with previous published results for other syncope detectors.

Comparison of Feature Selection Methods for Syncope Prediction

The aim of this study is to develop a method to predict unexplained syncope. Its diagnosis is currently based on the reproduction of symptoms induced by a 45-min of 60-80deg head-upright tilt test (HUTT). The main drawback of this test concerns its duration which can reach 45 minutes, therefore our study proposes an analysis which is only based on the 10 first minutes of the test. An important number of variables is obtained during the HUTT. To reduce and to select the most relevant variables, many feature selection methods are used and compared to obtain groups of pertinent variables. We used classification tools to achieve significant syncope outcome prediction.