Automated Diagnosis-Assisting Medical Devices Utilizing Pattern Localization Of Quasi-Periodic Signals

Abstract

A method for localizing a pattern in a quasi-periodic signal includes estimating, using a controller, a rate or a frequency of a quasi-periodic signal, and defining a search window based on the estimated rate or frequency of the quasi-periodic signal. A starting position is defined in the received quasi-periodic signal, the starting position corresponding to a first maximum. A portion of the quasi-periodic signal in the search window is cross-correlated with a template signal pattern to be matched to produce a second maximum. The second maximum is defined by the controller as a new starting position. The new starting position is stored.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This application claims the benefit of and priority to U.S. Provisional Patent Application No. 61 / 787,998, titled “Automated Diagnosis-Assisting Medical Devices Utilizing Rate / Frequency Estimation And Pattern Localization Of Quasi-Periodic Signals” and filed on Mar. 15, 2013, which is incorporated herein by reference in its respective entirety.COPYRIGHT[0002]A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent disclosure, as it appears in the Patent and Trademark Office patent files or records, but otherwise reserves all copyright rights whatsoever.FIELD OF THE INVENTION[0003]Various aspects of the present disclosure relate to the estimation of the rate or frequency and to the localization of similar patterns in quasi-periodic signals. More specifically, for example, the signals are not limi...

AI Technical Summary

Benefits of technology

[0011]Furthermore, yet another aspect of the present disclosure utilizes signal templates, e.g. a single representative heartbeat in a series of heartbeats or an analytical signal that shows similar features as the pattern of interest in the target signal, to search throughout the entire signal for locations where similar signal patters (or shapes) are found. The resulting locations are stored and returned. The algorithm utilizes a sequence of cross-correlation and windowing functions in combination with signal rate estimation that makes the algorithm robust against changes in periodicity, noise, and artefacts.

[0014]According to one attribute, the algorithm uses methods like auto-correlation or cross-correlation, which can be computed very efficiently by using time-frequency conversion to perform such operations. Microprocessors often provide optimized implementations of such time-frequency conversions, such as Fast Fourier Transform (“FFT”), and, therefore, significantly boosting time-domain operations.

[0015]According to another attribute, the algorithm enables fast and efficient computation by using pre-trained classifiers (e.g., neural networks, support vector machines, Bayesian networks, etc.). The pre-trained classifiers facilitate new data to be classified with simple and computationally efficient operations (e.g., matrix multiplications). For this approach, parameters are determined with training data. For example, in reference to neural networks, weights and biases determined with training data. Or, in reference to support vector machines, the location of the support vectors in the hyperspace is determined with training data. Comprehensive and well classified training data is useful for a good pre-training of classifiers. The training data of the disclosed algorithm includes, for example, raw phonocardiogram data and / or corresponding diagnoses (obtained using, for example, echocardiography as the gold standard method for diagnosing heart defects). With a comprehensive training set, a classifier can be optimized (or, pre-trained) and applied to new data, which enables fast and efficient computations. In contrast, so-called lazy-learning methods use the whole available data set (stored locally) and compare new data against the whole training set for classification. The lazy-learning methods lead to higher space requirements for storing the training data set and / or to increased computational costs for performing the classification.

[0016]According to yet another attribute, the features of interest (or inputs to the classifier) are determined in advance by feature selection algorithms (e.g., sequential floating forward selection), which reduce feature space. Features are also analyzed using statistical tools such as a principal component analysis, which results in linearly uncorrelated variables and which further optimizes the feature space. Hence, only the most powerful and meaningful features are selected for the algorithm, increasing its computational efficiency and robustness against noise and outliers.

[0023]According to other aspects of the present disclosure, the device does not require any external input from medical professionals or other devices (e.g., ECG), does not require traditional auscultation techniques to be modified, does not require especially quiet environments (such as, for example, the holding of breath by the patient during auscultation), and / or does not require manual or semi-automated analysis by a medical professional. Alternatively, adding an external input by the user is optional and does not hinder the device from performing its tasks. In fact, the external input might potentially even increase the accuracy of the results.

[0024]By way of example, in reference to a phonocardiogram analysis, the age of the patient is a helpful parameter for narrowing down the range of likely heart rates and possible diseases (e.g., a specific classification of a heart murmur). A newborn, for example, usually has a heart rate greater than 100 beats per minute and the range of possible diseases is generally different than, for example, for a child greater than 2 years of age. One or more features of the present disclosure are beneficial to and enhance existing electronic stethoscopes by increasing their function as a medical device and informing medical staff within a short period of time whether physiological signals are healthy or require further medical attention. Thus, one or more features of the present disclosure can be utilized directly on an electronic stethoscope or in combination with an electronic stethoscope and a portable device, wherein computations and interaction (e.g., visualization of the findings) with medical staff are achieved through the portable device.

Problems solved by technology

The signals are typically, but not necessarily limited to being quasi-periodic, and are often overlaid with noise or other artifacts.

Images